xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision 7cd22ac43418da08448d0bab1009ff3cbda85120)
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 
46 extern int newnfs_numnfsd;
47 extern struct nfsstatsv1 nfsstatsv1;
48 extern int nfsrv_lease;
49 extern struct timeval nfsboottime;
50 extern u_int32_t newnfs_true, newnfs_false;
51 extern struct mtx nfsrv_dslock_mtx;
52 extern struct mtx nfsrv_recalllock_mtx;
53 extern struct mtx nfsrv_dontlistlock_mtx;
54 extern int nfsd_debuglevel;
55 extern u_int nfsrv_dsdirsize;
56 extern struct nfsdevicehead nfsrv_devidhead;
57 extern int nfsrv_doflexfile;
58 extern int nfsrv_maxpnfsmirror;
59 NFSV4ROOTLOCKMUTEX;
60 NFSSTATESPINLOCK;
61 extern struct nfsdontlisthead nfsrv_dontlisthead;
62 extern volatile int nfsrv_devidcnt;
63 extern struct nfslayouthead nfsrv_recalllisthead;
64 extern char *nfsrv_zeropnfsdat;
65 
66 SYSCTL_DECL(_vfs_nfsd);
67 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
68 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
69     &nfsrv_statehashsize, 0,
70     "Size of state hash table set via loader.conf");
71 
72 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
73 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
74     &nfsrv_clienthashsize, 0,
75     "Size of client hash table set via loader.conf");
76 
77 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
78 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
79     &nfsrv_lockhashsize, 0,
80     "Size of file handle hash table set via loader.conf");
81 
82 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
83 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
84     &nfsrv_sessionhashsize, 0,
85     "Size of session hash table set via loader.conf");
86 
87 int	nfsrv_layouthighwater = NFSLAYOUTHIGHWATER;
88 SYSCTL_INT(_vfs_nfsd, OID_AUTO, layouthighwater, CTLFLAG_RDTUN,
89     &nfsrv_layouthighwater, 0,
90     "High water mark for number of layouts set via loader.conf");
91 
92 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
93 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
94     &nfsrv_v4statelimit, 0,
95     "High water limit for NFSv4 opens+locks+delegations");
96 
97 static int	nfsrv_writedelegifpos = 0;
98 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
99     &nfsrv_writedelegifpos, 0,
100     "Issue a write delegation for read opens if possible");
101 
102 static int	nfsrv_allowreadforwriteopen = 1;
103 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
104     &nfsrv_allowreadforwriteopen, 0,
105     "Allow Reads to be done with Write Access StateIDs");
106 
107 int	nfsrv_pnfsatime = 0;
108 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW,
109     &nfsrv_pnfsatime, 0,
110     "For pNFS service, do Getattr ops to keep atime up-to-date");
111 
112 int	nfsrv_flexlinuxhack = 0;
113 SYSCTL_INT(_vfs_nfsd, OID_AUTO, flexlinuxhack, CTLFLAG_RW,
114     &nfsrv_flexlinuxhack, 0,
115     "For Linux clients, hack around Flex File Layout bug");
116 
117 /*
118  * Hash lists for nfs V4.
119  */
120 struct nfsclienthashhead	*nfsclienthash;
121 struct nfslockhashhead		*nfslockhash;
122 struct nfssessionhash		*nfssessionhash;
123 struct nfslayouthash		*nfslayouthash;
124 volatile int nfsrv_dontlistlen = 0;
125 
126 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
127 static time_t nfsrvboottime;
128 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
129 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
130 static int nfsrv_nogsscallback = 0;
131 static volatile int nfsrv_writedelegcnt = 0;
132 static int nfsrv_faildscnt;
133 
134 /* local functions */
135 static void nfsrv_dumpaclient(struct nfsclient *clp,
136     struct nfsd_dumpclients *dumpp);
137 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
138     NFSPROC_T *p);
139 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
140     NFSPROC_T *p);
141 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
142     NFSPROC_T *p);
143 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
144     int cansleep, NFSPROC_T *p);
145 static void nfsrv_freenfslock(struct nfslock *lop);
146 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
147 static void nfsrv_freedeleg(struct nfsstate *);
148 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
149     u_int32_t flags, struct nfsstate **stpp);
150 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
151     struct nfsstate **stpp);
152 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
153     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
154 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
155     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
156 static void nfsrv_insertlock(struct nfslock *new_lop,
157     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
158 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
159     struct nfslock **other_lopp, struct nfslockfile *lfp);
160 static int nfsrv_getipnumber(u_char *cp);
161 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
162     nfsv4stateid_t *stateidp, int specialid);
163 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
164     u_int32_t flags);
165 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
166     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
167     struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p);
168 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
169     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
170     int *slotposp);
171 static u_int32_t nfsrv_nextclientindex(void);
172 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
173 static void nfsrv_markstable(struct nfsclient *clp);
174 static void nfsrv_markreclaim(struct nfsclient *clp);
175 static int nfsrv_checkstable(struct nfsclient *clp);
176 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
177     vnode *vp, NFSPROC_T *p);
178 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
179     NFSPROC_T *p, vnode_t vp);
180 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
181     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
182 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
183     struct nfsclient *clp);
184 static time_t nfsrv_leaseexpiry(void);
185 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
186 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
187     struct nfsstate *stp, struct nfsrvcache *op);
188 static int nfsrv_nootherstate(struct nfsstate *stp);
189 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
190     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
191 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
192     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
193 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
194     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
195     NFSPROC_T *p);
196 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
197     NFSPROC_T *p);
198 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
199     uint64_t first, uint64_t end);
200 static void nfsrv_locklf(struct nfslockfile *lfp);
201 static void nfsrv_unlocklf(struct nfslockfile *lfp);
202 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
203 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
204 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
205     int dont_replycache, struct nfsdsession **sepp, int *slotposp);
206 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
207 static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
208     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p);
209 static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp);
210 static void nfsrv_freelayoutlist(nfsquad_t clientid);
211 static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype,
212     int iomode);
213 static void nfsrv_freealllayouts(void);
214 static void nfsrv_freedevid(struct nfsdevice *ds);
215 static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
216     struct nfsdevice **dsp);
217 static void nfsrv_deleteds(struct nfsdevice *fndds);
218 static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost);
219 static void nfsrv_freealldevids(void);
220 static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp,
221     int maxcnt, NFSPROC_T *p);
222 static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp,
223     fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype,
224     NFSPROC_T *p);
225 static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
226     NFSPROC_T *, struct nfslayout **lypp);
227 static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt);
228 static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode,
229     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
230 static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode,
231     int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
232 static int nfsrv_dontlayout(fhandle_t *fhp);
233 static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
234     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
235     vnode_t *tvpp);
236 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp);
237 
238 /*
239  * Scan the client list for a match and either return the current one,
240  * create a new entry or return an error.
241  * If returning a non-error, the clp structure must either be linked into
242  * the client list or free'd.
243  */
244 int
245 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
246     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
247 {
248 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
249 	int i, error = 0, ret;
250 	struct nfsstate *stp, *tstp;
251 #ifdef INET
252 	struct sockaddr_in *sin, *rin;
253 #endif
254 #ifdef INET6
255 	struct sockaddr_in6 *sin6, *rin6;
256 #endif
257 	struct nfsdsession *sep, *nsep;
258 	int zapit = 0, gotit, hasstate = 0, igotlock;
259 	static u_int64_t confirm_index = 0;
260 
261 	/*
262 	 * Check for state resource limit exceeded.
263 	 */
264 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
265 		error = NFSERR_RESOURCE;
266 		goto out;
267 	}
268 
269 	if (nfsrv_issuedelegs == 0 ||
270 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
271 		/*
272 		 * Don't do callbacks when delegations are disabled or
273 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
274 		 * If establishing a callback connection is attempted
275 		 * when a firewall is blocking the callback path, the
276 		 * server may wait too long for the connect attempt to
277 		 * succeed during the Open. Some clients, such as Linux,
278 		 * may timeout and give up on the Open before the server
279 		 * replies. Also, since AUTH_GSS callbacks are not
280 		 * yet interoperability tested, they might cause the
281 		 * server to crap out, if they get past the Init call to
282 		 * the client.
283 		 */
284 		new_clp->lc_program = 0;
285 
286 	/* Lock out other nfsd threads */
287 	NFSLOCKV4ROOTMUTEX();
288 	nfsv4_relref(&nfsv4rootfs_lock);
289 	do {
290 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
291 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
292 	} while (!igotlock);
293 	NFSUNLOCKV4ROOTMUTEX();
294 
295 	/*
296 	 * Search for a match in the client list.
297 	 */
298 	gotit = i = 0;
299 	while (i < nfsrv_clienthashsize && !gotit) {
300 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
301 		if (new_clp->lc_idlen == clp->lc_idlen &&
302 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
303 			gotit = 1;
304 			break;
305 		}
306 	    }
307 	    if (gotit == 0)
308 		i++;
309 	}
310 	if (!gotit ||
311 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
312 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
313 			/*
314 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
315 			 * client is trying to update a confirmed clientid.
316 			 */
317 			NFSLOCKV4ROOTMUTEX();
318 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
319 			NFSUNLOCKV4ROOTMUTEX();
320 			confirmp->lval[1] = 0;
321 			error = NFSERR_NOENT;
322 			goto out;
323 		}
324 		/*
325 		 * Get rid of the old one.
326 		 */
327 		if (i != nfsrv_clienthashsize) {
328 			LIST_REMOVE(clp, lc_hash);
329 			nfsrv_cleanclient(clp, p);
330 			nfsrv_freedeleglist(&clp->lc_deleg);
331 			nfsrv_freedeleglist(&clp->lc_olddeleg);
332 			zapit = 1;
333 		}
334 		/*
335 		 * Add it after assigning a client id to it.
336 		 */
337 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
338 		if ((nd->nd_flag & ND_NFSV41) != 0)
339 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
340 			    ++confirm_index;
341 		else
342 			confirmp->qval = new_clp->lc_confirm.qval =
343 			    ++confirm_index;
344 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
345 		    (u_int32_t)nfsrvboottime;
346 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
347 		    nfsrv_nextclientindex();
348 		new_clp->lc_stateindex = 0;
349 		new_clp->lc_statemaxindex = 0;
350 		new_clp->lc_cbref = 0;
351 		new_clp->lc_expiry = nfsrv_leaseexpiry();
352 		LIST_INIT(&new_clp->lc_open);
353 		LIST_INIT(&new_clp->lc_deleg);
354 		LIST_INIT(&new_clp->lc_olddeleg);
355 		LIST_INIT(&new_clp->lc_session);
356 		for (i = 0; i < nfsrv_statehashsize; i++)
357 			LIST_INIT(&new_clp->lc_stateid[i]);
358 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
359 		    lc_hash);
360 		nfsstatsv1.srvclients++;
361 		nfsrv_openpluslock++;
362 		nfsrv_clients++;
363 		NFSLOCKV4ROOTMUTEX();
364 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
365 		NFSUNLOCKV4ROOTMUTEX();
366 		if (zapit)
367 			nfsrv_zapclient(clp, p);
368 		*new_clpp = NULL;
369 		goto out;
370 	}
371 
372 	/*
373 	 * Now, handle the cases where the id is already issued.
374 	 */
375 	if (nfsrv_notsamecredname(nd, clp)) {
376 	    /*
377 	     * Check to see if there is expired state that should go away.
378 	     */
379 	    if (clp->lc_expiry < NFSD_MONOSEC &&
380 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
381 		nfsrv_cleanclient(clp, p);
382 		nfsrv_freedeleglist(&clp->lc_deleg);
383 	    }
384 
385 	    /*
386 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
387 	     * RFC3530 Sec. 8.1.2 last para.
388 	     */
389 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
390 		hasstate = 1;
391 	    } else if (LIST_EMPTY(&clp->lc_open)) {
392 		hasstate = 0;
393 	    } else {
394 		hasstate = 0;
395 		/* Look for an Open on the OpenOwner */
396 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
397 		    if (!LIST_EMPTY(&stp->ls_open)) {
398 			hasstate = 1;
399 			break;
400 		    }
401 		}
402 	    }
403 	    if (hasstate) {
404 		/*
405 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
406 		 * filling out the correct ipaddr and portnum.
407 		 */
408 		switch (clp->lc_req.nr_nam->sa_family) {
409 #ifdef INET
410 		case AF_INET:
411 			sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam;
412 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
413 			sin->sin_addr.s_addr = rin->sin_addr.s_addr;
414 			sin->sin_port = rin->sin_port;
415 			break;
416 #endif
417 #ifdef INET6
418 		case AF_INET6:
419 			sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam;
420 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
421 			sin6->sin6_addr = rin6->sin6_addr;
422 			sin6->sin6_port = rin6->sin6_port;
423 			break;
424 #endif
425 		}
426 		NFSLOCKV4ROOTMUTEX();
427 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
428 		NFSUNLOCKV4ROOTMUTEX();
429 		error = NFSERR_CLIDINUSE;
430 		goto out;
431 	    }
432 	}
433 
434 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
435 		/*
436 		 * If the verifier has changed, the client has rebooted
437 		 * and a new client id is issued. The old state info
438 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
439 		 */
440 		LIST_REMOVE(clp, lc_hash);
441 
442 		/* Get rid of all sessions on this clientid. */
443 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) {
444 			ret = nfsrv_freesession(sep, NULL);
445 			if (ret != 0)
446 				printf("nfsrv_setclient: verifier changed free"
447 				    " session failed=%d\n", ret);
448 		}
449 
450 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
451 		if ((nd->nd_flag & ND_NFSV41) != 0)
452 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
453 			    ++confirm_index;
454 		else
455 			confirmp->qval = new_clp->lc_confirm.qval =
456 			    ++confirm_index;
457 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
458 		    nfsrvboottime;
459 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
460 		    nfsrv_nextclientindex();
461 		new_clp->lc_stateindex = 0;
462 		new_clp->lc_statemaxindex = 0;
463 		new_clp->lc_cbref = 0;
464 		new_clp->lc_expiry = nfsrv_leaseexpiry();
465 
466 		/*
467 		 * Save the state until confirmed.
468 		 */
469 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
470 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
471 			tstp->ls_clp = new_clp;
472 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
473 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
474 			tstp->ls_clp = new_clp;
475 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
476 		    ls_list);
477 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
478 			tstp->ls_clp = new_clp;
479 		for (i = 0; i < nfsrv_statehashsize; i++) {
480 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
481 			    &clp->lc_stateid[i], ls_hash);
482 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
483 				tstp->ls_clp = new_clp;
484 		}
485 		LIST_INIT(&new_clp->lc_session);
486 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
487 		    lc_hash);
488 		nfsstatsv1.srvclients++;
489 		nfsrv_openpluslock++;
490 		nfsrv_clients++;
491 		NFSLOCKV4ROOTMUTEX();
492 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
493 		NFSUNLOCKV4ROOTMUTEX();
494 
495 		/*
496 		 * Must wait until any outstanding callback on the old clp
497 		 * completes.
498 		 */
499 		NFSLOCKSTATE();
500 		while (clp->lc_cbref) {
501 			clp->lc_flags |= LCL_WAKEUPWANTED;
502 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
503 			    "nfsd clp", 10 * hz);
504 		}
505 		NFSUNLOCKSTATE();
506 		nfsrv_zapclient(clp, p);
507 		*new_clpp = NULL;
508 		goto out;
509 	}
510 
511 	/* For NFSv4.1, mark that we found a confirmed clientid. */
512 	if ((nd->nd_flag & ND_NFSV41) != 0) {
513 		clientidp->lval[0] = clp->lc_clientid.lval[0];
514 		clientidp->lval[1] = clp->lc_clientid.lval[1];
515 		confirmp->lval[0] = 0;	/* Ignored by client */
516 		confirmp->lval[1] = 1;
517 	} else {
518 		/*
519 		 * id and verifier match, so update the net address info
520 		 * and get rid of any existing callback authentication
521 		 * handle, so a new one will be acquired.
522 		 */
523 		LIST_REMOVE(clp, lc_hash);
524 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
525 		new_clp->lc_expiry = nfsrv_leaseexpiry();
526 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
527 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
528 		    clp->lc_clientid.lval[0];
529 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
530 		    clp->lc_clientid.lval[1];
531 		new_clp->lc_delegtime = clp->lc_delegtime;
532 		new_clp->lc_stateindex = clp->lc_stateindex;
533 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
534 		new_clp->lc_cbref = 0;
535 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
536 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
537 			tstp->ls_clp = new_clp;
538 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
539 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
540 			tstp->ls_clp = new_clp;
541 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
542 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
543 			tstp->ls_clp = new_clp;
544 		for (i = 0; i < nfsrv_statehashsize; i++) {
545 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
546 			    &clp->lc_stateid[i], ls_hash);
547 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
548 				tstp->ls_clp = new_clp;
549 		}
550 		LIST_INIT(&new_clp->lc_session);
551 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
552 		    lc_hash);
553 		nfsstatsv1.srvclients++;
554 		nfsrv_openpluslock++;
555 		nfsrv_clients++;
556 	}
557 	NFSLOCKV4ROOTMUTEX();
558 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
559 	NFSUNLOCKV4ROOTMUTEX();
560 
561 	if ((nd->nd_flag & ND_NFSV41) == 0) {
562 		/*
563 		 * Must wait until any outstanding callback on the old clp
564 		 * completes.
565 		 */
566 		NFSLOCKSTATE();
567 		while (clp->lc_cbref) {
568 			clp->lc_flags |= LCL_WAKEUPWANTED;
569 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
570 			    "nfsdclp", 10 * hz);
571 		}
572 		NFSUNLOCKSTATE();
573 		nfsrv_zapclient(clp, p);
574 		*new_clpp = NULL;
575 	}
576 
577 out:
578 	NFSEXITCODE2(error, nd);
579 	return (error);
580 }
581 
582 /*
583  * Check to see if the client id exists and optionally confirm it.
584  */
585 int
586 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
587     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
588     struct nfsrv_descript *nd, NFSPROC_T *p)
589 {
590 	struct nfsclient *clp;
591 	struct nfsstate *stp;
592 	int i;
593 	struct nfsclienthashhead *hp;
594 	int error = 0, igotlock, doneok;
595 	struct nfssessionhash *shp;
596 	struct nfsdsession *sep;
597 	uint64_t sessid[2];
598 	static uint64_t next_sess = 0;
599 
600 	if (clpp)
601 		*clpp = NULL;
602 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
603 	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
604 		error = NFSERR_STALECLIENTID;
605 		goto out;
606 	}
607 
608 	/*
609 	 * If called with opflags == CLOPS_RENEW, the State Lock is
610 	 * already held. Otherwise, we need to get either that or,
611 	 * for the case of Confirm, lock out the nfsd threads.
612 	 */
613 	if (opflags & CLOPS_CONFIRM) {
614 		NFSLOCKV4ROOTMUTEX();
615 		nfsv4_relref(&nfsv4rootfs_lock);
616 		do {
617 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
618 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
619 		} while (!igotlock);
620 		/*
621 		 * Create a new sessionid here, since we need to do it where
622 		 * there is a mutex held to serialize update of next_sess.
623 		 */
624 		if ((nd->nd_flag & ND_NFSV41) != 0) {
625 			sessid[0] = ++next_sess;
626 			sessid[1] = clientid.qval;
627 		}
628 		NFSUNLOCKV4ROOTMUTEX();
629 	} else if (opflags != CLOPS_RENEW) {
630 		NFSLOCKSTATE();
631 	}
632 
633 	/* For NFSv4.1, the clp is acquired from the associated session. */
634 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
635 	    opflags == CLOPS_RENEW) {
636 		clp = NULL;
637 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
638 			shp = NFSSESSIONHASH(nd->nd_sessionid);
639 			NFSLOCKSESSION(shp);
640 			sep = nfsrv_findsession(nd->nd_sessionid);
641 			if (sep != NULL)
642 				clp = sep->sess_clp;
643 			NFSUNLOCKSESSION(shp);
644 		}
645 	} else {
646 		hp = NFSCLIENTHASH(clientid);
647 		LIST_FOREACH(clp, hp, lc_hash) {
648 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
649 				break;
650 		}
651 	}
652 	if (clp == NULL) {
653 		if (opflags & CLOPS_CONFIRM)
654 			error = NFSERR_STALECLIENTID;
655 		else
656 			error = NFSERR_EXPIRED;
657 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
658 		/*
659 		 * If marked admin revoked, just return the error.
660 		 */
661 		error = NFSERR_ADMINREVOKED;
662 	}
663 	if (error) {
664 		if (opflags & CLOPS_CONFIRM) {
665 			NFSLOCKV4ROOTMUTEX();
666 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
667 			NFSUNLOCKV4ROOTMUTEX();
668 		} else if (opflags != CLOPS_RENEW) {
669 			NFSUNLOCKSTATE();
670 		}
671 		goto out;
672 	}
673 
674 	/*
675 	 * Perform any operations specified by the opflags.
676 	 */
677 	if (opflags & CLOPS_CONFIRM) {
678 		if (((nd->nd_flag & ND_NFSV41) != 0 &&
679 		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
680 		    ((nd->nd_flag & ND_NFSV41) == 0 &&
681 		     clp->lc_confirm.qval != confirm.qval))
682 			error = NFSERR_STALECLIENTID;
683 		else if (nfsrv_notsamecredname(nd, clp))
684 			error = NFSERR_CLIDINUSE;
685 
686 		if (!error) {
687 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
688 			LCL_NEEDSCONFIRM) {
689 			/*
690 			 * Hang onto the delegations (as old delegations)
691 			 * for an Open with CLAIM_DELEGATE_PREV unless in
692 			 * grace, but get rid of the rest of the state.
693 			 */
694 			nfsrv_cleanclient(clp, p);
695 			nfsrv_freedeleglist(&clp->lc_olddeleg);
696 			if (nfsrv_checkgrace(nd, clp, 0)) {
697 			    /* In grace, so just delete delegations */
698 			    nfsrv_freedeleglist(&clp->lc_deleg);
699 			} else {
700 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
701 				stp->ls_flags |= NFSLCK_OLDDELEG;
702 			    clp->lc_delegtime = NFSD_MONOSEC +
703 				nfsrv_lease + NFSRV_LEASEDELTA;
704 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
705 				ls_list);
706 			}
707 			if ((nd->nd_flag & ND_NFSV41) != 0)
708 			    clp->lc_program = cbprogram;
709 		    }
710 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
711 		    if (clp->lc_program)
712 			clp->lc_flags |= LCL_NEEDSCBNULL;
713 		    /* For NFSv4.1, link the session onto the client. */
714 		    if (nsep != NULL) {
715 			/* Hold a reference on the xprt for a backchannel. */
716 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
717 			    != 0) {
718 			    if (clp->lc_req.nr_client == NULL)
719 				clp->lc_req.nr_client = (struct __rpc_client *)
720 				    clnt_bck_create(nd->nd_xprt->xp_socket,
721 				    cbprogram, NFSV4_CBVERS);
722 			    if (clp->lc_req.nr_client != NULL) {
723 				SVC_ACQUIRE(nd->nd_xprt);
724 				CLNT_ACQUIRE(clp->lc_req.nr_client);
725 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
726 				/* Disable idle timeout. */
727 				nd->nd_xprt->xp_idletimeout = 0;
728 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
729 			    } else
730 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
731 			}
732 			NFSBCOPY(sessid, nsep->sess_sessionid,
733 			    NFSX_V4SESSIONID);
734 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
735 			    NFSX_V4SESSIONID);
736 			shp = NFSSESSIONHASH(nsep->sess_sessionid);
737 			NFSLOCKSTATE();
738 			NFSLOCKSESSION(shp);
739 			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
740 			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
741 			nsep->sess_clp = clp;
742 			NFSUNLOCKSESSION(shp);
743 			NFSUNLOCKSTATE();
744 		    }
745 		}
746 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
747 		error = NFSERR_EXPIRED;
748 	}
749 
750 	/*
751 	 * If called by the Renew Op, we must check the principal.
752 	 */
753 	if (!error && (opflags & CLOPS_RENEWOP)) {
754 	    if (nfsrv_notsamecredname(nd, clp)) {
755 		doneok = 0;
756 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
757 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
758 			if ((stp->ls_flags & NFSLCK_OPEN) &&
759 			    stp->ls_uid == nd->nd_cred->cr_uid) {
760 				doneok = 1;
761 				break;
762 			}
763 		    }
764 		}
765 		if (!doneok)
766 			error = NFSERR_ACCES;
767 	    }
768 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
769 		error = NFSERR_CBPATHDOWN;
770 	}
771 	if ((!error || error == NFSERR_CBPATHDOWN) &&
772 	     (opflags & CLOPS_RENEW)) {
773 		clp->lc_expiry = nfsrv_leaseexpiry();
774 	}
775 	if (opflags & CLOPS_CONFIRM) {
776 		NFSLOCKV4ROOTMUTEX();
777 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
778 		NFSUNLOCKV4ROOTMUTEX();
779 	} else if (opflags != CLOPS_RENEW) {
780 		NFSUNLOCKSTATE();
781 	}
782 	if (clpp)
783 		*clpp = clp;
784 
785 out:
786 	NFSEXITCODE2(error, nd);
787 	return (error);
788 }
789 
790 /*
791  * Perform the NFSv4.1 destroy clientid.
792  */
793 int
794 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
795 {
796 	struct nfsclient *clp;
797 	struct nfsclienthashhead *hp;
798 	int error = 0, i, igotlock;
799 
800 	if (nfsrvboottime != clientid.lval[0]) {
801 		error = NFSERR_STALECLIENTID;
802 		goto out;
803 	}
804 
805 	/* Lock out other nfsd threads */
806 	NFSLOCKV4ROOTMUTEX();
807 	nfsv4_relref(&nfsv4rootfs_lock);
808 	do {
809 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
810 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
811 	} while (igotlock == 0);
812 	NFSUNLOCKV4ROOTMUTEX();
813 
814 	hp = NFSCLIENTHASH(clientid);
815 	LIST_FOREACH(clp, hp, lc_hash) {
816 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
817 			break;
818 	}
819 	if (clp == NULL) {
820 		NFSLOCKV4ROOTMUTEX();
821 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
822 		NFSUNLOCKV4ROOTMUTEX();
823 		/* Just return ok, since it is gone. */
824 		goto out;
825 	}
826 
827 	/*
828 	 * Free up all layouts on the clientid.  Should the client return the
829 	 * layouts?
830 	 */
831 	nfsrv_freelayoutlist(clientid);
832 
833 	/* Scan for state on the clientid. */
834 	for (i = 0; i < nfsrv_statehashsize; i++)
835 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
836 			NFSLOCKV4ROOTMUTEX();
837 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
838 			NFSUNLOCKV4ROOTMUTEX();
839 			error = NFSERR_CLIENTIDBUSY;
840 			goto out;
841 		}
842 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
843 		NFSLOCKV4ROOTMUTEX();
844 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
845 		NFSUNLOCKV4ROOTMUTEX();
846 		error = NFSERR_CLIENTIDBUSY;
847 		goto out;
848 	}
849 
850 	/* Destroy the clientid and return ok. */
851 	nfsrv_cleanclient(clp, p);
852 	nfsrv_freedeleglist(&clp->lc_deleg);
853 	nfsrv_freedeleglist(&clp->lc_olddeleg);
854 	LIST_REMOVE(clp, lc_hash);
855 	NFSLOCKV4ROOTMUTEX();
856 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
857 	NFSUNLOCKV4ROOTMUTEX();
858 	nfsrv_zapclient(clp, p);
859 out:
860 	NFSEXITCODE2(error, nd);
861 	return (error);
862 }
863 
864 /*
865  * Called from the new nfssvc syscall to admin revoke a clientid.
866  * Returns 0 for success, error otherwise.
867  */
868 int
869 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
870 {
871 	struct nfsclient *clp = NULL;
872 	int i, error = 0;
873 	int gotit, igotlock;
874 
875 	/*
876 	 * First, lock out the nfsd so that state won't change while the
877 	 * revocation record is being written to the stable storage restart
878 	 * file.
879 	 */
880 	NFSLOCKV4ROOTMUTEX();
881 	do {
882 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
883 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
884 	} while (!igotlock);
885 	NFSUNLOCKV4ROOTMUTEX();
886 
887 	/*
888 	 * Search for a match in the client list.
889 	 */
890 	gotit = i = 0;
891 	while (i < nfsrv_clienthashsize && !gotit) {
892 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
893 		if (revokep->nclid_idlen == clp->lc_idlen &&
894 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
895 			gotit = 1;
896 			break;
897 		}
898 	    }
899 	    i++;
900 	}
901 	if (!gotit) {
902 		NFSLOCKV4ROOTMUTEX();
903 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
904 		NFSUNLOCKV4ROOTMUTEX();
905 		error = EPERM;
906 		goto out;
907 	}
908 
909 	/*
910 	 * Now, write out the revocation record
911 	 */
912 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
913 	nfsrv_backupstable();
914 
915 	/*
916 	 * and clear out the state, marking the clientid revoked.
917 	 */
918 	clp->lc_flags &= ~LCL_CALLBACKSON;
919 	clp->lc_flags |= LCL_ADMINREVOKED;
920 	nfsrv_cleanclient(clp, p);
921 	nfsrv_freedeleglist(&clp->lc_deleg);
922 	nfsrv_freedeleglist(&clp->lc_olddeleg);
923 	NFSLOCKV4ROOTMUTEX();
924 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
925 	NFSUNLOCKV4ROOTMUTEX();
926 
927 out:
928 	NFSEXITCODE(error);
929 	return (error);
930 }
931 
932 /*
933  * Dump out stats for all clients. Called from nfssvc(2), that is used
934  * nfsstatsv1.
935  */
936 void
937 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
938 {
939 	struct nfsclient *clp;
940 	int i = 0, cnt = 0;
941 
942 	/*
943 	 * First, get a reference on the nfsv4rootfs_lock so that an
944 	 * exclusive lock cannot be acquired while dumping the clients.
945 	 */
946 	NFSLOCKV4ROOTMUTEX();
947 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
948 	NFSUNLOCKV4ROOTMUTEX();
949 	NFSLOCKSTATE();
950 	/*
951 	 * Rattle through the client lists until done.
952 	 */
953 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
954 	    clp = LIST_FIRST(&nfsclienthash[i]);
955 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
956 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
957 		cnt++;
958 		clp = LIST_NEXT(clp, lc_hash);
959 	    }
960 	    i++;
961 	}
962 	if (cnt < maxcnt)
963 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
964 	NFSUNLOCKSTATE();
965 	NFSLOCKV4ROOTMUTEX();
966 	nfsv4_relref(&nfsv4rootfs_lock);
967 	NFSUNLOCKV4ROOTMUTEX();
968 }
969 
970 /*
971  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
972  */
973 static void
974 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
975 {
976 	struct nfsstate *stp, *openstp, *lckownstp;
977 	struct nfslock *lop;
978 	sa_family_t af;
979 #ifdef INET
980 	struct sockaddr_in *rin;
981 #endif
982 #ifdef INET6
983 	struct sockaddr_in6 *rin6;
984 #endif
985 
986 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
987 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
988 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
989 	dumpp->ndcl_flags = clp->lc_flags;
990 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
991 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
992 	af = clp->lc_req.nr_nam->sa_family;
993 	dumpp->ndcl_addrfam = af;
994 	switch (af) {
995 #ifdef INET
996 	case AF_INET:
997 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
998 		dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr;
999 		break;
1000 #endif
1001 #ifdef INET6
1002 	case AF_INET6:
1003 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
1004 		dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr;
1005 		break;
1006 #endif
1007 	}
1008 
1009 	/*
1010 	 * Now, scan the state lists and total up the opens and locks.
1011 	 */
1012 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
1013 	    dumpp->ndcl_nopenowners++;
1014 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
1015 		dumpp->ndcl_nopens++;
1016 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
1017 		    dumpp->ndcl_nlockowners++;
1018 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
1019 			dumpp->ndcl_nlocks++;
1020 		    }
1021 		}
1022 	    }
1023 	}
1024 
1025 	/*
1026 	 * and the delegation lists.
1027 	 */
1028 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
1029 	    dumpp->ndcl_ndelegs++;
1030 	}
1031 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
1032 	    dumpp->ndcl_nolddelegs++;
1033 	}
1034 }
1035 
1036 /*
1037  * Dump out lock stats for a file.
1038  */
1039 void
1040 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
1041     NFSPROC_T *p)
1042 {
1043 	struct nfsstate *stp;
1044 	struct nfslock *lop;
1045 	int cnt = 0;
1046 	struct nfslockfile *lfp;
1047 	sa_family_t af;
1048 #ifdef INET
1049 	struct sockaddr_in *rin;
1050 #endif
1051 #ifdef INET6
1052 	struct sockaddr_in6 *rin6;
1053 #endif
1054 	int ret;
1055 	fhandle_t nfh;
1056 
1057 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
1058 	/*
1059 	 * First, get a reference on the nfsv4rootfs_lock so that an
1060 	 * exclusive lock on it cannot be acquired while dumping the locks.
1061 	 */
1062 	NFSLOCKV4ROOTMUTEX();
1063 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1064 	NFSUNLOCKV4ROOTMUTEX();
1065 	NFSLOCKSTATE();
1066 	if (!ret)
1067 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1068 	if (ret) {
1069 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
1070 		NFSUNLOCKSTATE();
1071 		NFSLOCKV4ROOTMUTEX();
1072 		nfsv4_relref(&nfsv4rootfs_lock);
1073 		NFSUNLOCKV4ROOTMUTEX();
1074 		return;
1075 	}
1076 
1077 	/*
1078 	 * For each open share on file, dump it out.
1079 	 */
1080 	stp = LIST_FIRST(&lfp->lf_open);
1081 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1082 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1083 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1084 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1085 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1086 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1087 		ldumpp[cnt].ndlck_owner.nclid_idlen =
1088 		    stp->ls_openowner->ls_ownerlen;
1089 		NFSBCOPY(stp->ls_openowner->ls_owner,
1090 		    ldumpp[cnt].ndlck_owner.nclid_id,
1091 		    stp->ls_openowner->ls_ownerlen);
1092 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1093 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1094 		    stp->ls_clp->lc_idlen);
1095 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1096 		ldumpp[cnt].ndlck_addrfam = af;
1097 		switch (af) {
1098 #ifdef INET
1099 		case AF_INET:
1100 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1101 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1102 			break;
1103 #endif
1104 #ifdef INET6
1105 		case AF_INET6:
1106 			rin6 = (struct sockaddr_in6 *)
1107 			    stp->ls_clp->lc_req.nr_nam;
1108 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1109 			break;
1110 #endif
1111 		}
1112 		stp = LIST_NEXT(stp, ls_file);
1113 		cnt++;
1114 	}
1115 
1116 	/*
1117 	 * and all locks.
1118 	 */
1119 	lop = LIST_FIRST(&lfp->lf_lock);
1120 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1121 		stp = lop->lo_stp;
1122 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1123 		ldumpp[cnt].ndlck_first = lop->lo_first;
1124 		ldumpp[cnt].ndlck_end = lop->lo_end;
1125 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1126 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1127 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1128 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1129 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1130 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1131 		    stp->ls_ownerlen);
1132 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1133 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1134 		    stp->ls_clp->lc_idlen);
1135 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1136 		ldumpp[cnt].ndlck_addrfam = af;
1137 		switch (af) {
1138 #ifdef INET
1139 		case AF_INET:
1140 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1141 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1142 			break;
1143 #endif
1144 #ifdef INET6
1145 		case AF_INET6:
1146 			rin6 = (struct sockaddr_in6 *)
1147 			    stp->ls_clp->lc_req.nr_nam;
1148 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1149 			break;
1150 #endif
1151 		}
1152 		lop = LIST_NEXT(lop, lo_lckfile);
1153 		cnt++;
1154 	}
1155 
1156 	/*
1157 	 * and the delegations.
1158 	 */
1159 	stp = LIST_FIRST(&lfp->lf_deleg);
1160 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1161 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1162 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1163 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1164 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1165 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1166 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1167 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1168 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1169 		    stp->ls_clp->lc_idlen);
1170 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1171 		ldumpp[cnt].ndlck_addrfam = af;
1172 		switch (af) {
1173 #ifdef INET
1174 		case AF_INET:
1175 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1176 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1177 			break;
1178 #endif
1179 #ifdef INET6
1180 		case AF_INET6:
1181 			rin6 = (struct sockaddr_in6 *)
1182 			    stp->ls_clp->lc_req.nr_nam;
1183 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1184 			break;
1185 #endif
1186 		}
1187 		stp = LIST_NEXT(stp, ls_file);
1188 		cnt++;
1189 	}
1190 
1191 	/*
1192 	 * If list isn't full, mark end of list by setting the client name
1193 	 * to zero length.
1194 	 */
1195 	if (cnt < maxcnt)
1196 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1197 	NFSUNLOCKSTATE();
1198 	NFSLOCKV4ROOTMUTEX();
1199 	nfsv4_relref(&nfsv4rootfs_lock);
1200 	NFSUNLOCKV4ROOTMUTEX();
1201 }
1202 
1203 /*
1204  * Server timer routine. It can scan any linked list, so long
1205  * as it holds the spin/mutex lock and there is no exclusive lock on
1206  * nfsv4rootfs_lock.
1207  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1208  *  to do this from a callout, since the spin locks work. For
1209  *  Darwin, I'm not sure what will work correctly yet.)
1210  * Should be called once per second.
1211  */
1212 void
1213 nfsrv_servertimer(void)
1214 {
1215 	struct nfsclient *clp, *nclp;
1216 	struct nfsstate *stp, *nstp;
1217 	int got_ref, i;
1218 
1219 	/*
1220 	 * Make sure nfsboottime is set. This is used by V3 as well
1221 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1222 	 * only used by the V4 server for leases.
1223 	 */
1224 	if (nfsboottime.tv_sec == 0)
1225 		NFSSETBOOTTIME(nfsboottime);
1226 
1227 	/*
1228 	 * If server hasn't started yet, just return.
1229 	 */
1230 	NFSLOCKSTATE();
1231 	if (nfsrv_stablefirst.nsf_eograce == 0) {
1232 		NFSUNLOCKSTATE();
1233 		return;
1234 	}
1235 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1236 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1237 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1238 			nfsrv_stablefirst.nsf_flags |=
1239 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1240 		NFSUNLOCKSTATE();
1241 		return;
1242 	}
1243 
1244 	/*
1245 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1246 	 * no nfsd thread can acquire an exclusive lock on it before this
1247 	 * call is done. If it is already exclusively locked, just return.
1248 	 */
1249 	NFSLOCKV4ROOTMUTEX();
1250 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1251 	NFSUNLOCKV4ROOTMUTEX();
1252 	if (got_ref == 0) {
1253 		NFSUNLOCKSTATE();
1254 		return;
1255 	}
1256 
1257 	/*
1258 	 * For each client...
1259 	 */
1260 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1261 	    clp = LIST_FIRST(&nfsclienthash[i]);
1262 	    while (clp != LIST_END(&nfsclienthash[i])) {
1263 		nclp = LIST_NEXT(clp, lc_hash);
1264 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1265 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1266 			 && ((LIST_EMPTY(&clp->lc_deleg)
1267 			      && LIST_EMPTY(&clp->lc_open)) ||
1268 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1269 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1270 			(clp->lc_expiry < NFSD_MONOSEC &&
1271 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1272 			/*
1273 			 * Lease has expired several nfsrv_lease times ago:
1274 			 * PLUS
1275 			 *    - no state is associated with it
1276 			 *    OR
1277 			 *    - above high water mark for number of clients
1278 			 *      (nfsrv_clienthighwater should be large enough
1279 			 *       that this only occurs when clients fail to
1280 			 *       use the same nfs_client_id4.id. Maybe somewhat
1281 			 *       higher that the maximum number of clients that
1282 			 *       will mount this server?)
1283 			 * OR
1284 			 * Lease has expired a very long time ago
1285 			 * OR
1286 			 * Lease has expired PLUS the number of opens + locks
1287 			 * has exceeded 90% of capacity
1288 			 *
1289 			 * --> Mark for expiry. The actual expiry will be done
1290 			 *     by an nfsd sometime soon.
1291 			 */
1292 			clp->lc_flags |= LCL_EXPIREIT;
1293 			nfsrv_stablefirst.nsf_flags |=
1294 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1295 		    } else {
1296 			/*
1297 			 * If there are no opens, increment no open tick cnt
1298 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1299 			 * otherwise, if there is an open, reset no open time
1300 			 * Hopefully, this will avoid excessive re-creation
1301 			 * of open owners and subsequent open confirms.
1302 			 */
1303 			stp = LIST_FIRST(&clp->lc_open);
1304 			while (stp != LIST_END(&clp->lc_open)) {
1305 				nstp = LIST_NEXT(stp, ls_list);
1306 				if (LIST_EMPTY(&stp->ls_open)) {
1307 					stp->ls_noopens++;
1308 					if (stp->ls_noopens > NFSNOOPEN ||
1309 					    (nfsrv_openpluslock * 2) >
1310 					    nfsrv_v4statelimit)
1311 						nfsrv_stablefirst.nsf_flags |=
1312 							NFSNSF_NOOPENS;
1313 				} else {
1314 					stp->ls_noopens = 0;
1315 				}
1316 				stp = nstp;
1317 			}
1318 		    }
1319 		}
1320 		clp = nclp;
1321 	    }
1322 	}
1323 	NFSUNLOCKSTATE();
1324 	NFSLOCKV4ROOTMUTEX();
1325 	nfsv4_relref(&nfsv4rootfs_lock);
1326 	NFSUNLOCKV4ROOTMUTEX();
1327 }
1328 
1329 /*
1330  * The following set of functions free up the various data structures.
1331  */
1332 /*
1333  * Clear out all open/lock state related to this nfsclient.
1334  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1335  * there are no other active nfsd threads.
1336  */
1337 void
1338 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1339 {
1340 	struct nfsstate *stp, *nstp;
1341 	struct nfsdsession *sep, *nsep;
1342 
1343 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1344 		nfsrv_freeopenowner(stp, 1, p);
1345 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1346 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1347 			(void)nfsrv_freesession(sep, NULL);
1348 }
1349 
1350 /*
1351  * Free a client that has been cleaned. It should also already have been
1352  * removed from the lists.
1353  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1354  *  softclock interrupts are enabled.)
1355  */
1356 void
1357 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1358 {
1359 
1360 #ifdef notyet
1361 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1362 	     (LCL_GSS | LCL_CALLBACKSON) &&
1363 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1364 	    clp->lc_handlelen > 0) {
1365 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1366 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1367 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1368 			NULL, 0, NULL, NULL, NULL, 0, p);
1369 	}
1370 #endif
1371 	newnfs_disconnect(&clp->lc_req);
1372 	free(clp->lc_req.nr_nam, M_SONAME);
1373 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1374 	free(clp->lc_stateid, M_NFSDCLIENT);
1375 	free(clp, M_NFSDCLIENT);
1376 	NFSLOCKSTATE();
1377 	nfsstatsv1.srvclients--;
1378 	nfsrv_openpluslock--;
1379 	nfsrv_clients--;
1380 	NFSUNLOCKSTATE();
1381 }
1382 
1383 /*
1384  * Free a list of delegation state structures.
1385  * (This function will also free all nfslockfile structures that no
1386  *  longer have associated state.)
1387  */
1388 void
1389 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1390 {
1391 	struct nfsstate *stp, *nstp;
1392 
1393 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1394 		nfsrv_freedeleg(stp);
1395 	}
1396 	LIST_INIT(sthp);
1397 }
1398 
1399 /*
1400  * Free up a delegation.
1401  */
1402 static void
1403 nfsrv_freedeleg(struct nfsstate *stp)
1404 {
1405 	struct nfslockfile *lfp;
1406 
1407 	LIST_REMOVE(stp, ls_hash);
1408 	LIST_REMOVE(stp, ls_list);
1409 	LIST_REMOVE(stp, ls_file);
1410 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1411 		nfsrv_writedelegcnt--;
1412 	lfp = stp->ls_lfp;
1413 	if (LIST_EMPTY(&lfp->lf_open) &&
1414 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1415 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1416 	    lfp->lf_usecount == 0 &&
1417 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1418 		nfsrv_freenfslockfile(lfp);
1419 	free(stp, M_NFSDSTATE);
1420 	nfsstatsv1.srvdelegates--;
1421 	nfsrv_openpluslock--;
1422 	nfsrv_delegatecnt--;
1423 }
1424 
1425 /*
1426  * This function frees an open owner and all associated opens.
1427  */
1428 static void
1429 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1430 {
1431 	struct nfsstate *nstp, *tstp;
1432 
1433 	LIST_REMOVE(stp, ls_list);
1434 	/*
1435 	 * Now, free all associated opens.
1436 	 */
1437 	nstp = LIST_FIRST(&stp->ls_open);
1438 	while (nstp != LIST_END(&stp->ls_open)) {
1439 		tstp = nstp;
1440 		nstp = LIST_NEXT(nstp, ls_list);
1441 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1442 	}
1443 	if (stp->ls_op)
1444 		nfsrvd_derefcache(stp->ls_op);
1445 	free(stp, M_NFSDSTATE);
1446 	nfsstatsv1.srvopenowners--;
1447 	nfsrv_openpluslock--;
1448 }
1449 
1450 /*
1451  * This function frees an open (nfsstate open structure) with all associated
1452  * lock_owners and locks. It also frees the nfslockfile structure iff there
1453  * are no other opens on the file.
1454  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1455  */
1456 static int
1457 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1458 {
1459 	struct nfsstate *nstp, *tstp;
1460 	struct nfslockfile *lfp;
1461 	int ret;
1462 
1463 	LIST_REMOVE(stp, ls_hash);
1464 	LIST_REMOVE(stp, ls_list);
1465 	LIST_REMOVE(stp, ls_file);
1466 
1467 	lfp = stp->ls_lfp;
1468 	/*
1469 	 * Now, free all lockowners associated with this open.
1470 	 */
1471 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1472 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1473 
1474 	/*
1475 	 * The nfslockfile is freed here if there are no locks
1476 	 * associated with the open.
1477 	 * If there are locks associated with the open, the
1478 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1479 	 * Acquire the state mutex to avoid races with calls to
1480 	 * nfsrv_getlockfile().
1481 	 */
1482 	if (cansleep != 0)
1483 		NFSLOCKSTATE();
1484 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1485 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1486 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1487 	    lfp->lf_usecount == 0 &&
1488 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1489 		nfsrv_freenfslockfile(lfp);
1490 		ret = 1;
1491 	} else
1492 		ret = 0;
1493 	if (cansleep != 0)
1494 		NFSUNLOCKSTATE();
1495 	free(stp, M_NFSDSTATE);
1496 	nfsstatsv1.srvopens--;
1497 	nfsrv_openpluslock--;
1498 	return (ret);
1499 }
1500 
1501 /*
1502  * Frees a lockowner and all associated locks.
1503  */
1504 static void
1505 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1506     NFSPROC_T *p)
1507 {
1508 
1509 	LIST_REMOVE(stp, ls_hash);
1510 	LIST_REMOVE(stp, ls_list);
1511 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1512 	if (stp->ls_op)
1513 		nfsrvd_derefcache(stp->ls_op);
1514 	free(stp, M_NFSDSTATE);
1515 	nfsstatsv1.srvlockowners--;
1516 	nfsrv_openpluslock--;
1517 }
1518 
1519 /*
1520  * Free all the nfs locks on a lockowner.
1521  */
1522 static void
1523 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1524     NFSPROC_T *p)
1525 {
1526 	struct nfslock *lop, *nlop;
1527 	struct nfsrollback *rlp, *nrlp;
1528 	struct nfslockfile *lfp = NULL;
1529 	int gottvp = 0;
1530 	vnode_t tvp = NULL;
1531 	uint64_t first, end;
1532 
1533 	if (vp != NULL)
1534 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1535 	lop = LIST_FIRST(&stp->ls_lock);
1536 	while (lop != LIST_END(&stp->ls_lock)) {
1537 		nlop = LIST_NEXT(lop, lo_lckowner);
1538 		/*
1539 		 * Since all locks should be for the same file, lfp should
1540 		 * not change.
1541 		 */
1542 		if (lfp == NULL)
1543 			lfp = lop->lo_lfp;
1544 		else if (lfp != lop->lo_lfp)
1545 			panic("allnfslocks");
1546 		/*
1547 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1548 		 * from the file handle. This only occurs when called from
1549 		 * nfsrv_cleanclient().
1550 		 */
1551 		if (gottvp == 0) {
1552 			if (nfsrv_dolocallocks == 0)
1553 				tvp = NULL;
1554 			else if (vp == NULL && cansleep != 0) {
1555 				tvp = nfsvno_getvp(&lfp->lf_fh);
1556 				if (tvp != NULL)
1557 					NFSVOPUNLOCK(tvp);
1558 			} else
1559 				tvp = vp;
1560 			gottvp = 1;
1561 		}
1562 
1563 		if (tvp != NULL) {
1564 			if (cansleep == 0)
1565 				panic("allnfs2");
1566 			first = lop->lo_first;
1567 			end = lop->lo_end;
1568 			nfsrv_freenfslock(lop);
1569 			nfsrv_localunlock(tvp, lfp, first, end, p);
1570 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1571 			    nrlp)
1572 				free(rlp, M_NFSDROLLBACK);
1573 			LIST_INIT(&lfp->lf_rollback);
1574 		} else
1575 			nfsrv_freenfslock(lop);
1576 		lop = nlop;
1577 	}
1578 	if (vp == NULL && tvp != NULL)
1579 		vrele(tvp);
1580 }
1581 
1582 /*
1583  * Free an nfslock structure.
1584  */
1585 static void
1586 nfsrv_freenfslock(struct nfslock *lop)
1587 {
1588 
1589 	if (lop->lo_lckfile.le_prev != NULL) {
1590 		LIST_REMOVE(lop, lo_lckfile);
1591 		nfsstatsv1.srvlocks--;
1592 		nfsrv_openpluslock--;
1593 	}
1594 	LIST_REMOVE(lop, lo_lckowner);
1595 	free(lop, M_NFSDLOCK);
1596 }
1597 
1598 /*
1599  * This function frees an nfslockfile structure.
1600  */
1601 static void
1602 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1603 {
1604 
1605 	LIST_REMOVE(lfp, lf_hash);
1606 	free(lfp, M_NFSDLOCKFILE);
1607 }
1608 
1609 /*
1610  * This function looks up an nfsstate structure via stateid.
1611  */
1612 static int
1613 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1614     struct nfsstate **stpp)
1615 {
1616 	struct nfsstate *stp;
1617 	struct nfsstatehead *hp;
1618 	int error = 0;
1619 
1620 	*stpp = NULL;
1621 	hp = NFSSTATEHASH(clp, *stateidp);
1622 	LIST_FOREACH(stp, hp, ls_hash) {
1623 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1624 			NFSX_STATEIDOTHER))
1625 			break;
1626 	}
1627 
1628 	/*
1629 	 * If no state id in list, return NFSERR_BADSTATEID.
1630 	 */
1631 	if (stp == LIST_END(hp)) {
1632 		error = NFSERR_BADSTATEID;
1633 		goto out;
1634 	}
1635 	*stpp = stp;
1636 
1637 out:
1638 	NFSEXITCODE(error);
1639 	return (error);
1640 }
1641 
1642 /*
1643  * This function gets an nfsstate structure via owner string.
1644  */
1645 static void
1646 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1647     struct nfsstate **stpp)
1648 {
1649 	struct nfsstate *stp;
1650 
1651 	*stpp = NULL;
1652 	LIST_FOREACH(stp, hp, ls_list) {
1653 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1654 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1655 			*stpp = stp;
1656 			return;
1657 		}
1658 	}
1659 }
1660 
1661 /*
1662  * Lock control function called to update lock status.
1663  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1664  * that one isn't to be created and an NFSERR_xxx for other errors.
1665  * The structures new_stp and new_lop are passed in as pointers that should
1666  * be set to NULL if the structure is used and shouldn't be free'd.
1667  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1668  * never used and can safely be allocated on the stack. For all other
1669  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1670  * in case they are used.
1671  */
1672 int
1673 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1674     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1675     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1676     __unused struct nfsexstuff *exp,
1677     struct nfsrv_descript *nd, NFSPROC_T *p)
1678 {
1679 	struct nfslock *lop;
1680 	struct nfsstate *new_stp = *new_stpp;
1681 	struct nfslock *new_lop = *new_lopp;
1682 	struct nfsstate *tstp, *mystp, *nstp;
1683 	int specialid = 0;
1684 	struct nfslockfile *lfp;
1685 	struct nfslock *other_lop = NULL;
1686 	struct nfsstate *stp, *lckstp = NULL;
1687 	struct nfsclient *clp = NULL;
1688 	u_int32_t bits;
1689 	int error = 0, haslock = 0, ret, reterr;
1690 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1691 	fhandle_t nfh;
1692 	uint64_t first, end;
1693 	uint32_t lock_flags;
1694 
1695 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1696 		/*
1697 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1698 		 * let reads with all 1s go ahead.
1699 		 */
1700 		if (new_stp->ls_stateid.seqid == 0x0 &&
1701 		    new_stp->ls_stateid.other[0] == 0x0 &&
1702 		    new_stp->ls_stateid.other[1] == 0x0 &&
1703 		    new_stp->ls_stateid.other[2] == 0x0)
1704 			specialid = 1;
1705 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1706 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1707 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1708 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1709 			specialid = 2;
1710 	}
1711 
1712 	/*
1713 	 * Check for restart conditions (client and server).
1714 	 */
1715 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1716 	    &new_stp->ls_stateid, specialid);
1717 	if (error)
1718 		goto out;
1719 
1720 	/*
1721 	 * Check for state resource limit exceeded.
1722 	 */
1723 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1724 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1725 		error = NFSERR_RESOURCE;
1726 		goto out;
1727 	}
1728 
1729 	/*
1730 	 * For the lock case, get another nfslock structure,
1731 	 * just in case we need it.
1732 	 * Malloc now, before we start sifting through the linked lists,
1733 	 * in case we have to wait for memory.
1734 	 */
1735 tryagain:
1736 	if (new_stp->ls_flags & NFSLCK_LOCK)
1737 		other_lop = malloc(sizeof (struct nfslock),
1738 		    M_NFSDLOCK, M_WAITOK);
1739 	filestruct_locked = 0;
1740 	reterr = 0;
1741 	lfp = NULL;
1742 
1743 	/*
1744 	 * Get the lockfile structure for CFH now, so we can do a sanity
1745 	 * check against the stateid, before incrementing the seqid#, since
1746 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1747 	 * shouldn't be incremented for this case.
1748 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1749 	 * will be handled later.
1750 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1751 	 * lock the nfslockfile structure.
1752 	 */
1753 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1754 	NFSLOCKSTATE();
1755 	if (getlckret == 0) {
1756 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1757 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1758 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1759 			    &lfp, &nfh, 1);
1760 			if (getlckret == 0)
1761 				filestruct_locked = 1;
1762 		} else
1763 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1764 			    &lfp, &nfh, 0);
1765 	}
1766 	if (getlckret != 0 && getlckret != -1)
1767 		reterr = getlckret;
1768 
1769 	if (filestruct_locked != 0) {
1770 		LIST_INIT(&lfp->lf_rollback);
1771 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1772 			/*
1773 			 * For local locking, do the advisory locking now, so
1774 			 * that any conflict can be detected. A failure later
1775 			 * can be rolled back locally. If an error is returned,
1776 			 * struct nfslockfile has been unlocked and any local
1777 			 * locking rolled back.
1778 			 */
1779 			NFSUNLOCKSTATE();
1780 			if (vnode_unlocked == 0) {
1781 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1782 				vnode_unlocked = 1;
1783 				NFSVOPUNLOCK(vp);
1784 			}
1785 			reterr = nfsrv_locallock(vp, lfp,
1786 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1787 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1788 			NFSLOCKSTATE();
1789 		}
1790 	}
1791 
1792 	if (specialid == 0) {
1793 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1794 		/*
1795 		 * RFC 3530 does not list LockT as an op that renews a
1796 		 * lease, but the consensus seems to be that it is ok
1797 		 * for a server to do so.
1798 		 */
1799 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1800 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1801 
1802 		/*
1803 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1804 		 * error returns for LockT, just go ahead and test for a lock,
1805 		 * since there are no locks for this client, but other locks
1806 		 * can conflict. (ie. same client will always be false)
1807 		 */
1808 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1809 		    error = 0;
1810 		lckstp = new_stp;
1811 	    } else {
1812 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1813 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1814 	      if (error == 0)
1815 		/*
1816 		 * Look up the stateid
1817 		 */
1818 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1819 		  new_stp->ls_flags, &stp);
1820 	      /*
1821 	       * do some sanity checks for an unconfirmed open or a
1822 	       * stateid that refers to the wrong file, for an open stateid
1823 	       */
1824 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1825 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1826 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1827 		      /*
1828 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1829 		       * The only exception is using SETATTR with SIZE.
1830 		       * */
1831                     if ((new_stp->ls_flags &
1832                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1833 			     error = NFSERR_BADSTATEID;
1834 	      }
1835 
1836 		if (error == 0 &&
1837 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1838 		  getlckret == 0 && stp->ls_lfp != lfp)
1839 			error = NFSERR_BADSTATEID;
1840 
1841 	      /*
1842 	       * If the lockowner stateid doesn't refer to the same file,
1843 	       * I believe that is considered ok, since some clients will
1844 	       * only create a single lockowner and use that for all locks
1845 	       * on all files.
1846 	       * For now, log it as a diagnostic, instead of considering it
1847 	       * a BadStateid.
1848 	       */
1849 	      if (error == 0 && (stp->ls_flags &
1850 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1851 		  getlckret == 0 && stp->ls_lfp != lfp) {
1852 #ifdef DIAGNOSTIC
1853 		  printf("Got a lock statid for different file open\n");
1854 #endif
1855 		  /*
1856 		  error = NFSERR_BADSTATEID;
1857 		  */
1858 	      }
1859 
1860 	      if (error == 0) {
1861 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1862 			/*
1863 			 * If haslock set, we've already checked the seqid.
1864 			 */
1865 			if (!haslock) {
1866 			    if (stp->ls_flags & NFSLCK_OPEN)
1867 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1868 				    stp->ls_openowner, new_stp->ls_op);
1869 			    else
1870 				error = NFSERR_BADSTATEID;
1871 			}
1872 			if (!error)
1873 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1874 			if (lckstp) {
1875 			    /*
1876 			     * For NFSv4.1 and NFSv4.2 allow an
1877 			     * open_to_lock_owner when the lock_owner already
1878 			     * exists.  Just clear NFSLCK_OPENTOLOCK so that
1879 			     * a new lock_owner will not be created.
1880 			     * RFC7530 states that the error for NFSv4.0
1881 			     * is NFS4ERR_BAD_SEQID.
1882 			     */
1883 			    if ((nd->nd_flag & ND_NFSV41) != 0)
1884 				new_stp->ls_flags &= ~NFSLCK_OPENTOLOCK;
1885 			    else
1886 				error = NFSERR_BADSEQID;
1887 			} else
1888 			    lckstp = new_stp;
1889 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1890 			/*
1891 			 * If haslock set, ditto above.
1892 			 */
1893 			if (!haslock) {
1894 			    if (stp->ls_flags & NFSLCK_OPEN)
1895 				error = NFSERR_BADSTATEID;
1896 			    else
1897 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1898 				    stp, new_stp->ls_op);
1899 			}
1900 			lckstp = stp;
1901 		    } else {
1902 			lckstp = stp;
1903 		    }
1904 	      }
1905 	      /*
1906 	       * If the seqid part of the stateid isn't the same, return
1907 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1908 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1909 	       * nfsrv_returnoldstateid is set. (The consensus on the email
1910 	       * list was that most clients would prefer to not receive
1911 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1912 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1913 	       * allow for either server configuration.)
1914 	       */
1915 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1916 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1917 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1918 		    nfsrv_returnoldstateid)) ||
1919 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1920 		    new_stp->ls_stateid.seqid != 0)))
1921 		    error = NFSERR_OLDSTATEID;
1922 	    }
1923 	}
1924 
1925 	/*
1926 	 * Now we can check for grace.
1927 	 */
1928 	if (!error)
1929 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1930 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1931 		nfsrv_checkstable(clp))
1932 		error = NFSERR_NOGRACE;
1933 	/*
1934 	 * If we successfully Reclaimed state, note that.
1935 	 */
1936 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1937 		nfsrv_markstable(clp);
1938 
1939 	/*
1940 	 * At this point, either error == NFSERR_BADSTATEID or the
1941 	 * seqid# has been updated, so we can return any error.
1942 	 * If error == 0, there may be an error in:
1943 	 *    nd_repstat - Set by the calling function.
1944 	 *    reterr - Set above, if getting the nfslockfile structure
1945 	 *       or acquiring the local lock failed.
1946 	 *    (If both of these are set, nd_repstat should probably be
1947 	 *     returned, since that error was detected before this
1948 	 *     function call.)
1949 	 */
1950 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1951 		if (error == 0) {
1952 			if (nd->nd_repstat != 0)
1953 				error = nd->nd_repstat;
1954 			else
1955 				error = reterr;
1956 		}
1957 		if (filestruct_locked != 0) {
1958 			/* Roll back local locks. */
1959 			NFSUNLOCKSTATE();
1960 			if (vnode_unlocked == 0) {
1961 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1962 				vnode_unlocked = 1;
1963 				NFSVOPUNLOCK(vp);
1964 			}
1965 			nfsrv_locallock_rollback(vp, lfp, p);
1966 			NFSLOCKSTATE();
1967 			nfsrv_unlocklf(lfp);
1968 		}
1969 		NFSUNLOCKSTATE();
1970 		goto out;
1971 	}
1972 
1973 	/*
1974 	 * Check the nfsrv_getlockfile return.
1975 	 * Returned -1 if no structure found.
1976 	 */
1977 	if (getlckret == -1) {
1978 		error = NFSERR_EXPIRED;
1979 		/*
1980 		 * Called from lockt, so no lock is OK.
1981 		 */
1982 		if (new_stp->ls_flags & NFSLCK_TEST) {
1983 			error = 0;
1984 		} else if (new_stp->ls_flags &
1985 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1986 			/*
1987 			 * Called to check for a lock, OK if the stateid is all
1988 			 * 1s or all 0s, but there should be an nfsstate
1989 			 * otherwise.
1990 			 * (ie. If there is no open, I'll assume no share
1991 			 *  deny bits.)
1992 			 */
1993 			if (specialid)
1994 				error = 0;
1995 			else
1996 				error = NFSERR_BADSTATEID;
1997 		}
1998 		NFSUNLOCKSTATE();
1999 		goto out;
2000 	}
2001 
2002 	/*
2003 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
2004 	 * For NFSLCK_CHECK, allow a read if write access is granted,
2005 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
2006 	 * which implies a conflicting deny can't exist.
2007 	 */
2008 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
2009 	    /*
2010 	     * Four kinds of state id:
2011 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
2012 	     * - stateid for an open
2013 	     * - stateid for a delegation
2014 	     * - stateid for a lock owner
2015 	     */
2016 	    if (!specialid) {
2017 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2018 		    delegation = 1;
2019 		    mystp = stp;
2020 		    nfsrv_delaydelegtimeout(stp);
2021 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
2022 		    mystp = stp;
2023 		} else {
2024 		    mystp = stp->ls_openstp;
2025 		}
2026 		/*
2027 		 * If locking or checking, require correct access
2028 		 * bit set.
2029 		 */
2030 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
2031 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
2032 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
2033 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
2034 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
2035 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
2036 		     nfsrv_allowreadforwriteopen == 0) ||
2037 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
2038 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
2039 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
2040 			if (filestruct_locked != 0) {
2041 				/* Roll back local locks. */
2042 				NFSUNLOCKSTATE();
2043 				if (vnode_unlocked == 0) {
2044 					ASSERT_VOP_ELOCKED(vp,
2045 					    "nfsrv_lockctrl3");
2046 					vnode_unlocked = 1;
2047 					NFSVOPUNLOCK(vp);
2048 				}
2049 				nfsrv_locallock_rollback(vp, lfp, p);
2050 				NFSLOCKSTATE();
2051 				nfsrv_unlocklf(lfp);
2052 			}
2053 			NFSUNLOCKSTATE();
2054 			error = NFSERR_OPENMODE;
2055 			goto out;
2056 		}
2057 	    } else
2058 		mystp = NULL;
2059 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
2060 		/*
2061 		 * Check for a conflicting deny bit.
2062 		 */
2063 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
2064 		    if (tstp != mystp) {
2065 			bits = tstp->ls_flags;
2066 			bits >>= NFSLCK_SHIFT;
2067 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
2068 			    KASSERT(vnode_unlocked == 0,
2069 				("nfsrv_lockctrl: vnode unlocked1"));
2070 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
2071 				vp, p);
2072 			    if (ret == 1) {
2073 				/*
2074 				* nfsrv_clientconflict unlocks state
2075 				 * when it returns non-zero.
2076 				 */
2077 				lckstp = NULL;
2078 				goto tryagain;
2079 			    }
2080 			    if (ret == 0)
2081 				NFSUNLOCKSTATE();
2082 			    if (ret == 2)
2083 				error = NFSERR_PERM;
2084 			    else
2085 				error = NFSERR_OPENMODE;
2086 			    goto out;
2087 			}
2088 		    }
2089 		}
2090 
2091 		/* We're outta here */
2092 		NFSUNLOCKSTATE();
2093 		goto out;
2094 	    }
2095 	}
2096 
2097 	/*
2098 	 * For setattr, just get rid of all the Delegations for other clients.
2099 	 */
2100 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
2101 		KASSERT(vnode_unlocked == 0,
2102 		    ("nfsrv_lockctrl: vnode unlocked2"));
2103 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2104 		if (ret) {
2105 			/*
2106 			 * nfsrv_cleandeleg() unlocks state when it
2107 			 * returns non-zero.
2108 			 */
2109 			if (ret == -1) {
2110 				lckstp = NULL;
2111 				goto tryagain;
2112 			}
2113 			error = ret;
2114 			goto out;
2115 		}
2116 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2117 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2118 		     LIST_EMPTY(&lfp->lf_deleg))) {
2119 			NFSUNLOCKSTATE();
2120 			goto out;
2121 		}
2122 	}
2123 
2124 	/*
2125 	 * Check for a conflicting delegation. If one is found, call
2126 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2127 	 * been set yet, it will get the lock. Otherwise, it will recall
2128 	 * the delegation. Then, we try try again...
2129 	 * I currently believe the conflict algorithm to be:
2130 	 * For Lock Ops (Lock/LockT/LockU)
2131 	 * - there is a conflict iff a different client has a write delegation
2132 	 * For Reading (Read Op)
2133 	 * - there is a conflict iff a different client has a write delegation
2134 	 *   (the specialids are always a different client)
2135 	 * For Writing (Write/Setattr of size)
2136 	 * - there is a conflict if a different client has any delegation
2137 	 * - there is a conflict if the same client has a read delegation
2138 	 *   (I don't understand why this isn't allowed, but that seems to be
2139 	 *    the current consensus?)
2140 	 */
2141 	tstp = LIST_FIRST(&lfp->lf_deleg);
2142 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2143 	    nstp = LIST_NEXT(tstp, ls_file);
2144 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2145 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2146 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2147 		  clp != tstp->ls_clp &&
2148 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2149 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2150 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2151 		  (clp != tstp->ls_clp ||
2152 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2153 		ret = 0;
2154 		if (filestruct_locked != 0) {
2155 			/* Roll back local locks. */
2156 			NFSUNLOCKSTATE();
2157 			if (vnode_unlocked == 0) {
2158 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2159 				NFSVOPUNLOCK(vp);
2160 			}
2161 			nfsrv_locallock_rollback(vp, lfp, p);
2162 			NFSLOCKSTATE();
2163 			nfsrv_unlocklf(lfp);
2164 			NFSUNLOCKSTATE();
2165 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2166 			vnode_unlocked = 0;
2167 			if (VN_IS_DOOMED(vp))
2168 				ret = NFSERR_SERVERFAULT;
2169 			NFSLOCKSTATE();
2170 		}
2171 		if (ret == 0)
2172 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2173 		if (ret) {
2174 		    /*
2175 		     * nfsrv_delegconflict unlocks state when it
2176 		     * returns non-zero, which it always does.
2177 		     */
2178 		    if (other_lop) {
2179 			free(other_lop, M_NFSDLOCK);
2180 			other_lop = NULL;
2181 		    }
2182 		    if (ret == -1) {
2183 			lckstp = NULL;
2184 			goto tryagain;
2185 		    }
2186 		    error = ret;
2187 		    goto out;
2188 		}
2189 		/* Never gets here. */
2190 	    }
2191 	    tstp = nstp;
2192 	}
2193 
2194 	/*
2195 	 * Handle the unlock case by calling nfsrv_updatelock().
2196 	 * (Should I have done some access checking above for unlock? For now,
2197 	 *  just let it happen.)
2198 	 */
2199 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2200 		first = new_lop->lo_first;
2201 		end = new_lop->lo_end;
2202 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2203 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2204 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2205 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2206 		stateidp->other[0] = stp->ls_stateid.other[0];
2207 		stateidp->other[1] = stp->ls_stateid.other[1];
2208 		stateidp->other[2] = stp->ls_stateid.other[2];
2209 		if (filestruct_locked != 0) {
2210 			NFSUNLOCKSTATE();
2211 			if (vnode_unlocked == 0) {
2212 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2213 				vnode_unlocked = 1;
2214 				NFSVOPUNLOCK(vp);
2215 			}
2216 			/* Update the local locks. */
2217 			nfsrv_localunlock(vp, lfp, first, end, p);
2218 			NFSLOCKSTATE();
2219 			nfsrv_unlocklf(lfp);
2220 		}
2221 		NFSUNLOCKSTATE();
2222 		goto out;
2223 	}
2224 
2225 	/*
2226 	 * Search for a conflicting lock. A lock conflicts if:
2227 	 * - the lock range overlaps and
2228 	 * - at least one lock is a write lock and
2229 	 * - it is not owned by the same lock owner
2230 	 */
2231 	if (!delegation) {
2232 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2233 	    if (new_lop->lo_end > lop->lo_first &&
2234 		new_lop->lo_first < lop->lo_end &&
2235 		(new_lop->lo_flags == NFSLCK_WRITE ||
2236 		 lop->lo_flags == NFSLCK_WRITE) &&
2237 		lckstp != lop->lo_stp &&
2238 		(clp != lop->lo_stp->ls_clp ||
2239 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2240 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2241 		    lckstp->ls_ownerlen))) {
2242 		if (other_lop) {
2243 		    free(other_lop, M_NFSDLOCK);
2244 		    other_lop = NULL;
2245 		}
2246 		if (vnode_unlocked != 0)
2247 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2248 			NULL, p);
2249 		else
2250 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2251 			vp, p);
2252 		if (ret == 1) {
2253 		    if (filestruct_locked != 0) {
2254 			if (vnode_unlocked == 0) {
2255 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2256 				NFSVOPUNLOCK(vp);
2257 			}
2258 			/* Roll back local locks. */
2259 			nfsrv_locallock_rollback(vp, lfp, p);
2260 			NFSLOCKSTATE();
2261 			nfsrv_unlocklf(lfp);
2262 			NFSUNLOCKSTATE();
2263 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2264 			vnode_unlocked = 0;
2265 			if (VN_IS_DOOMED(vp)) {
2266 				error = NFSERR_SERVERFAULT;
2267 				goto out;
2268 			}
2269 		    }
2270 		    /*
2271 		     * nfsrv_clientconflict() unlocks state when it
2272 		     * returns non-zero.
2273 		     */
2274 		    lckstp = NULL;
2275 		    goto tryagain;
2276 		}
2277 		/*
2278 		 * Found a conflicting lock, so record the conflict and
2279 		 * return the error.
2280 		 */
2281 		if (cfp != NULL && ret == 0) {
2282 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2283 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2284 		    cfp->cl_first = lop->lo_first;
2285 		    cfp->cl_end = lop->lo_end;
2286 		    cfp->cl_flags = lop->lo_flags;
2287 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2288 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2289 			cfp->cl_ownerlen);
2290 		}
2291 		if (ret == 2)
2292 		    error = NFSERR_PERM;
2293 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2294 		    error = NFSERR_RECLAIMCONFLICT;
2295 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2296 		    error = NFSERR_LOCKED;
2297 		else
2298 		    error = NFSERR_DENIED;
2299 		if (filestruct_locked != 0 && ret == 0) {
2300 			/* Roll back local locks. */
2301 			NFSUNLOCKSTATE();
2302 			if (vnode_unlocked == 0) {
2303 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2304 				vnode_unlocked = 1;
2305 				NFSVOPUNLOCK(vp);
2306 			}
2307 			nfsrv_locallock_rollback(vp, lfp, p);
2308 			NFSLOCKSTATE();
2309 			nfsrv_unlocklf(lfp);
2310 		}
2311 		if (ret == 0)
2312 			NFSUNLOCKSTATE();
2313 		goto out;
2314 	    }
2315 	  }
2316 	}
2317 
2318 	/*
2319 	 * We only get here if there was no lock that conflicted.
2320 	 */
2321 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2322 		NFSUNLOCKSTATE();
2323 		goto out;
2324 	}
2325 
2326 	/*
2327 	 * We only get here when we are creating or modifying a lock.
2328 	 * There are two variants:
2329 	 * - exist_lock_owner where lock_owner exists
2330 	 * - open_to_lock_owner with new lock_owner
2331 	 */
2332 	first = new_lop->lo_first;
2333 	end = new_lop->lo_end;
2334 	lock_flags = new_lop->lo_flags;
2335 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2336 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2337 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2338 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2339 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2340 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2341 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2342 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2343 	} else {
2344 		/*
2345 		 * The new open_to_lock_owner case.
2346 		 * Link the new nfsstate into the lists.
2347 		 */
2348 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2349 		nfsrvd_refcache(new_stp->ls_op);
2350 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2351 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2352 		    clp->lc_clientid.lval[0];
2353 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2354 		    clp->lc_clientid.lval[1];
2355 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2356 		    nfsrv_nextstateindex(clp);
2357 		new_stp->ls_clp = clp;
2358 		LIST_INIT(&new_stp->ls_lock);
2359 		new_stp->ls_openstp = stp;
2360 		new_stp->ls_lfp = lfp;
2361 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2362 		    lfp);
2363 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2364 		    new_stp, ls_hash);
2365 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2366 		*new_lopp = NULL;
2367 		*new_stpp = NULL;
2368 		nfsstatsv1.srvlockowners++;
2369 		nfsrv_openpluslock++;
2370 	}
2371 	if (filestruct_locked != 0) {
2372 		NFSUNLOCKSTATE();
2373 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2374 		NFSLOCKSTATE();
2375 		nfsrv_unlocklf(lfp);
2376 	}
2377 	NFSUNLOCKSTATE();
2378 
2379 out:
2380 	if (haslock) {
2381 		NFSLOCKV4ROOTMUTEX();
2382 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2383 		NFSUNLOCKV4ROOTMUTEX();
2384 	}
2385 	if (vnode_unlocked != 0) {
2386 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2387 		if (error == 0 && VN_IS_DOOMED(vp))
2388 			error = NFSERR_SERVERFAULT;
2389 	}
2390 	if (other_lop)
2391 		free(other_lop, M_NFSDLOCK);
2392 	NFSEXITCODE2(error, nd);
2393 	return (error);
2394 }
2395 
2396 /*
2397  * Check for state errors for Open.
2398  * repstat is passed back out as an error if more critical errors
2399  * are not detected.
2400  */
2401 int
2402 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2403     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2404     NFSPROC_T *p, int repstat)
2405 {
2406 	struct nfsstate *stp, *nstp;
2407 	struct nfsclient *clp;
2408 	struct nfsstate *ownerstp;
2409 	struct nfslockfile *lfp, *new_lfp;
2410 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2411 
2412 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2413 		readonly = 1;
2414 	/*
2415 	 * Check for restart conditions (client and server).
2416 	 */
2417 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2418 		&new_stp->ls_stateid, 0);
2419 	if (error)
2420 		goto out;
2421 
2422 	/*
2423 	 * Check for state resource limit exceeded.
2424 	 * Technically this should be SMP protected, but the worst
2425 	 * case error is "out by one or two" on the count when it
2426 	 * returns NFSERR_RESOURCE and the limit is just a rather
2427 	 * arbitrary high water mark, so no harm is done.
2428 	 */
2429 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2430 		error = NFSERR_RESOURCE;
2431 		goto out;
2432 	}
2433 
2434 tryagain:
2435 	new_lfp = malloc(sizeof (struct nfslockfile),
2436 	    M_NFSDLOCKFILE, M_WAITOK);
2437 	if (vp)
2438 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2439 		    NULL, p);
2440 	NFSLOCKSTATE();
2441 	/*
2442 	 * Get the nfsclient structure.
2443 	 */
2444 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2445 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2446 
2447 	/*
2448 	 * Look up the open owner. See if it needs confirmation and
2449 	 * check the seq#, as required.
2450 	 */
2451 	if (!error)
2452 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2453 
2454 	if (!error && ownerstp) {
2455 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2456 		    new_stp->ls_op);
2457 		/*
2458 		 * If the OpenOwner hasn't been confirmed, assume the
2459 		 * old one was a replay and this one is ok.
2460 		 * See: RFC3530 Sec. 14.2.18.
2461 		 */
2462 		if (error == NFSERR_BADSEQID &&
2463 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2464 			error = 0;
2465 	}
2466 
2467 	/*
2468 	 * Check for grace.
2469 	 */
2470 	if (!error)
2471 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2472 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2473 		nfsrv_checkstable(clp))
2474 		error = NFSERR_NOGRACE;
2475 
2476 	/*
2477 	 * If none of the above errors occurred, let repstat be
2478 	 * returned.
2479 	 */
2480 	if (repstat && !error)
2481 		error = repstat;
2482 	if (error) {
2483 		NFSUNLOCKSTATE();
2484 		if (haslock) {
2485 			NFSLOCKV4ROOTMUTEX();
2486 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2487 			NFSUNLOCKV4ROOTMUTEX();
2488 		}
2489 		free(new_lfp, M_NFSDLOCKFILE);
2490 		goto out;
2491 	}
2492 
2493 	/*
2494 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2495 	 * (This always happens on the first pass, so haslock must be 0.)
2496 	 */
2497 	if (vp == NULL) {
2498 		NFSUNLOCKSTATE();
2499 		free(new_lfp, M_NFSDLOCKFILE);
2500 		goto out;
2501 	}
2502 
2503 	/*
2504 	 * Get the structure for the underlying file.
2505 	 */
2506 	if (getfhret)
2507 		error = getfhret;
2508 	else
2509 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2510 		    NULL, 0);
2511 	if (new_lfp)
2512 		free(new_lfp, M_NFSDLOCKFILE);
2513 	if (error) {
2514 		NFSUNLOCKSTATE();
2515 		if (haslock) {
2516 			NFSLOCKV4ROOTMUTEX();
2517 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2518 			NFSUNLOCKV4ROOTMUTEX();
2519 		}
2520 		goto out;
2521 	}
2522 
2523 	/*
2524 	 * Search for a conflicting open/share.
2525 	 */
2526 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2527 	    /*
2528 	     * For Delegate_Cur, search for the matching Delegation,
2529 	     * which indicates no conflict.
2530 	     * An old delegation should have been recovered by the
2531 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2532 	     * it match and return NFSERR_EXPIRED. Should I let it
2533 	     * match?
2534 	     */
2535 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2536 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2537 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2538 		    stateidp->seqid == 0) ||
2539 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2540 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2541 			  NFSX_STATEIDOTHER))
2542 			break;
2543 	    }
2544 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2545 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2546 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2547 		NFSUNLOCKSTATE();
2548 		if (haslock) {
2549 			NFSLOCKV4ROOTMUTEX();
2550 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2551 			NFSUNLOCKV4ROOTMUTEX();
2552 		}
2553 		error = NFSERR_EXPIRED;
2554 		goto out;
2555 	    }
2556 	}
2557 
2558 	/*
2559 	 * Check for access/deny bit conflicts. I check for the same
2560 	 * owner as well, in case the client didn't bother.
2561 	 */
2562 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2563 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2564 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2565 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2566 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2567 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2568 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2569 			if (ret == 1) {
2570 				/*
2571 				 * nfsrv_clientconflict() unlocks
2572 				 * state when it returns non-zero.
2573 				 */
2574 				goto tryagain;
2575 			}
2576 			if (ret == 2)
2577 				error = NFSERR_PERM;
2578 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2579 				error = NFSERR_RECLAIMCONFLICT;
2580 			else
2581 				error = NFSERR_SHAREDENIED;
2582 			if (ret == 0)
2583 				NFSUNLOCKSTATE();
2584 			if (haslock) {
2585 				NFSLOCKV4ROOTMUTEX();
2586 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2587 				NFSUNLOCKV4ROOTMUTEX();
2588 			}
2589 			goto out;
2590 		}
2591 	}
2592 
2593 	/*
2594 	 * Check for a conflicting delegation. If one is found, call
2595 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2596 	 * been set yet, it will get the lock. Otherwise, it will recall
2597 	 * the delegation. Then, we try try again...
2598 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2599 	 *  isn't a conflict.)
2600 	 * I currently believe the conflict algorithm to be:
2601 	 * For Open with Read Access and Deny None
2602 	 * - there is a conflict iff a different client has a write delegation
2603 	 * For Open with other Write Access or any Deny except None
2604 	 * - there is a conflict if a different client has any delegation
2605 	 * - there is a conflict if the same client has a read delegation
2606 	 *   (The current consensus is that this last case should be
2607 	 *    considered a conflict since the client with a read delegation
2608 	 *    could have done an Open with ReadAccess and WriteDeny
2609 	 *    locally and then not have checked for the WriteDeny.)
2610 	 * Don't check for a Reclaim, since that will be dealt with
2611 	 * by nfsrv_openctrl().
2612 	 */
2613 	if (!(new_stp->ls_flags &
2614 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2615 	    stp = LIST_FIRST(&lfp->lf_deleg);
2616 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2617 		nstp = LIST_NEXT(stp, ls_file);
2618 		if ((readonly && stp->ls_clp != clp &&
2619 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2620 		    (!readonly && (stp->ls_clp != clp ||
2621 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2622 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2623 			if (ret) {
2624 			    /*
2625 			     * nfsrv_delegconflict() unlocks state
2626 			     * when it returns non-zero.
2627 			     */
2628 			    if (ret == -1)
2629 				goto tryagain;
2630 			    error = ret;
2631 			    goto out;
2632 			}
2633 		}
2634 		stp = nstp;
2635 	    }
2636 	}
2637 	NFSUNLOCKSTATE();
2638 	if (haslock) {
2639 		NFSLOCKV4ROOTMUTEX();
2640 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2641 		NFSUNLOCKV4ROOTMUTEX();
2642 	}
2643 
2644 out:
2645 	NFSEXITCODE2(error, nd);
2646 	return (error);
2647 }
2648 
2649 /*
2650  * Open control function to create/update open state for an open.
2651  */
2652 int
2653 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2654     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2655     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2656     NFSPROC_T *p, u_quad_t filerev)
2657 {
2658 	struct nfsstate *new_stp = *new_stpp;
2659 	struct nfsstate *stp, *nstp;
2660 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2661 	struct nfslockfile *lfp, *new_lfp;
2662 	struct nfsclient *clp;
2663 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2664 	int readonly = 0, cbret = 1, getfhret = 0;
2665 	int gotstate = 0, len = 0;
2666 	u_char *clidp = NULL;
2667 
2668 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2669 		readonly = 1;
2670 	/*
2671 	 * Check for restart conditions (client and server).
2672 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2673 	 * If an error does show up, return NFSERR_EXPIRED, since the
2674 	 * the seqid# has already been incremented.
2675 	 */
2676 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2677 	    &new_stp->ls_stateid, 0);
2678 	if (error) {
2679 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2680 		    error);
2681 		error = NFSERR_EXPIRED;
2682 		goto out;
2683 	}
2684 
2685 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2686 tryagain:
2687 	new_lfp = malloc(sizeof (struct nfslockfile),
2688 	    M_NFSDLOCKFILE, M_WAITOK);
2689 	new_open = malloc(sizeof (struct nfsstate),
2690 	    M_NFSDSTATE, M_WAITOK);
2691 	new_deleg = malloc(sizeof (struct nfsstate),
2692 	    M_NFSDSTATE, M_WAITOK);
2693 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2694 	    NULL, p);
2695 	NFSLOCKSTATE();
2696 	/*
2697 	 * Get the client structure. Since the linked lists could be changed
2698 	 * by other nfsd processes if this process does a tsleep(), one of
2699 	 * two things must be done.
2700 	 * 1 - don't tsleep()
2701 	 * or
2702 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2703 	 *     before using the lists, since this lock stops the other
2704 	 *     nfsd. This should only be used for rare cases, since it
2705 	 *     essentially single threads the nfsd.
2706 	 *     At this time, it is only done for cases where the stable
2707 	 *     storage file must be written prior to completion of state
2708 	 *     expiration.
2709 	 */
2710 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2711 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2712 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2713 	    clp->lc_program) {
2714 		/*
2715 		 * This happens on the first open for a client
2716 		 * that supports callbacks.
2717 		 */
2718 		NFSUNLOCKSTATE();
2719 		/*
2720 		 * Although nfsrv_docallback() will sleep, clp won't
2721 		 * go away, since they are only removed when the
2722 		 * nfsv4_lock() has blocked the nfsd threads. The
2723 		 * fields in clp can change, but having multiple
2724 		 * threads do this Null callback RPC should be
2725 		 * harmless.
2726 		 */
2727 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2728 		    NULL, 0, NULL, NULL, NULL, 0, p);
2729 		NFSLOCKSTATE();
2730 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2731 		if (!cbret)
2732 			clp->lc_flags |= LCL_CALLBACKSON;
2733 	}
2734 
2735 	/*
2736 	 * Look up the open owner. See if it needs confirmation and
2737 	 * check the seq#, as required.
2738 	 */
2739 	if (!error)
2740 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2741 
2742 	if (error) {
2743 		NFSUNLOCKSTATE();
2744 		printf("Nfsd: openctrl unexpected state err=%d\n",
2745 			error);
2746 		free(new_lfp, M_NFSDLOCKFILE);
2747 		free(new_open, M_NFSDSTATE);
2748 		free(new_deleg, M_NFSDSTATE);
2749 		if (haslock) {
2750 			NFSLOCKV4ROOTMUTEX();
2751 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2752 			NFSUNLOCKV4ROOTMUTEX();
2753 		}
2754 		error = NFSERR_EXPIRED;
2755 		goto out;
2756 	}
2757 
2758 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2759 		nfsrv_markstable(clp);
2760 
2761 	/*
2762 	 * Get the structure for the underlying file.
2763 	 */
2764 	if (getfhret)
2765 		error = getfhret;
2766 	else
2767 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2768 		    NULL, 0);
2769 	if (new_lfp)
2770 		free(new_lfp, M_NFSDLOCKFILE);
2771 	if (error) {
2772 		NFSUNLOCKSTATE();
2773 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2774 		    error);
2775 		free(new_open, M_NFSDSTATE);
2776 		free(new_deleg, M_NFSDSTATE);
2777 		if (haslock) {
2778 			NFSLOCKV4ROOTMUTEX();
2779 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2780 			NFSUNLOCKV4ROOTMUTEX();
2781 		}
2782 		goto out;
2783 	}
2784 
2785 	/*
2786 	 * Search for a conflicting open/share.
2787 	 */
2788 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2789 	    /*
2790 	     * For Delegate_Cur, search for the matching Delegation,
2791 	     * which indicates no conflict.
2792 	     * An old delegation should have been recovered by the
2793 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2794 	     * it match and return NFSERR_EXPIRED. Should I let it
2795 	     * match?
2796 	     */
2797 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2798 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2799 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2800 		    stateidp->seqid == 0) ||
2801 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2802 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2803 			NFSX_STATEIDOTHER))
2804 			break;
2805 	    }
2806 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2807 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2808 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2809 		NFSUNLOCKSTATE();
2810 		printf("Nfsd openctrl unexpected expiry\n");
2811 		free(new_open, M_NFSDSTATE);
2812 		free(new_deleg, M_NFSDSTATE);
2813 		if (haslock) {
2814 			NFSLOCKV4ROOTMUTEX();
2815 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2816 			NFSUNLOCKV4ROOTMUTEX();
2817 		}
2818 		error = NFSERR_EXPIRED;
2819 		goto out;
2820 	    }
2821 
2822 	    /*
2823 	     * Don't issue a Delegation, since one already exists and
2824 	     * delay delegation timeout, as required.
2825 	     */
2826 	    delegate = 0;
2827 	    nfsrv_delaydelegtimeout(stp);
2828 	}
2829 
2830 	/*
2831 	 * Check for access/deny bit conflicts. I also check for the
2832 	 * same owner, since the client might not have bothered to check.
2833 	 * Also, note an open for the same file and owner, if found,
2834 	 * which is all we do here for Delegate_Cur, since conflict
2835 	 * checking is already done.
2836 	 */
2837 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2838 		if (ownerstp && stp->ls_openowner == ownerstp)
2839 			openstp = stp;
2840 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2841 		    /*
2842 		     * If another client has the file open, the only
2843 		     * delegation that can be issued is a Read delegation
2844 		     * and only if it is a Read open with Deny none.
2845 		     */
2846 		    if (clp != stp->ls_clp) {
2847 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2848 			    NFSLCK_READACCESS)
2849 			    writedeleg = 0;
2850 			else
2851 			    delegate = 0;
2852 		    }
2853 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2854 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2855 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2856 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2857 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2858 			if (ret == 1) {
2859 				/*
2860 				 * nfsrv_clientconflict() unlocks state
2861 				 * when it returns non-zero.
2862 				 */
2863 				free(new_open, M_NFSDSTATE);
2864 				free(new_deleg, M_NFSDSTATE);
2865 				openstp = NULL;
2866 				goto tryagain;
2867 			}
2868 			if (ret == 2)
2869 				error = NFSERR_PERM;
2870 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2871 				error = NFSERR_RECLAIMCONFLICT;
2872 			else
2873 				error = NFSERR_SHAREDENIED;
2874 			if (ret == 0)
2875 				NFSUNLOCKSTATE();
2876 			if (haslock) {
2877 				NFSLOCKV4ROOTMUTEX();
2878 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2879 				NFSUNLOCKV4ROOTMUTEX();
2880 			}
2881 			free(new_open, M_NFSDSTATE);
2882 			free(new_deleg, M_NFSDSTATE);
2883 			printf("nfsd openctrl unexpected client cnfl\n");
2884 			goto out;
2885 		    }
2886 		}
2887 	}
2888 
2889 	/*
2890 	 * Check for a conflicting delegation. If one is found, call
2891 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2892 	 * been set yet, it will get the lock. Otherwise, it will recall
2893 	 * the delegation. Then, we try try again...
2894 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2895 	 *  isn't a conflict.)
2896 	 * I currently believe the conflict algorithm to be:
2897 	 * For Open with Read Access and Deny None
2898 	 * - there is a conflict iff a different client has a write delegation
2899 	 * For Open with other Write Access or any Deny except None
2900 	 * - there is a conflict if a different client has any delegation
2901 	 * - there is a conflict if the same client has a read delegation
2902 	 *   (The current consensus is that this last case should be
2903 	 *    considered a conflict since the client with a read delegation
2904 	 *    could have done an Open with ReadAccess and WriteDeny
2905 	 *    locally and then not have checked for the WriteDeny.)
2906 	 */
2907 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2908 	    stp = LIST_FIRST(&lfp->lf_deleg);
2909 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2910 		nstp = LIST_NEXT(stp, ls_file);
2911 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2912 			writedeleg = 0;
2913 		else
2914 			delegate = 0;
2915 		if ((readonly && stp->ls_clp != clp &&
2916 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2917 		    (!readonly && (stp->ls_clp != clp ||
2918 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2919 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2920 			delegate = 2;
2921 		    } else {
2922 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2923 			if (ret) {
2924 			    /*
2925 			     * nfsrv_delegconflict() unlocks state
2926 			     * when it returns non-zero.
2927 			     */
2928 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2929 			    free(new_open, M_NFSDSTATE);
2930 			    free(new_deleg, M_NFSDSTATE);
2931 			    if (ret == -1) {
2932 				openstp = NULL;
2933 				goto tryagain;
2934 			    }
2935 			    error = ret;
2936 			    goto out;
2937 			}
2938 		    }
2939 		}
2940 		stp = nstp;
2941 	    }
2942 	}
2943 
2944 	/*
2945 	 * We only get here if there was no open that conflicted.
2946 	 * If an open for the owner exists, or in the access/deny bits.
2947 	 * Otherwise it is a new open. If the open_owner hasn't been
2948 	 * confirmed, replace the open with the new one needing confirmation,
2949 	 * otherwise add the open.
2950 	 */
2951 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2952 	    /*
2953 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2954 	     * a match. If found, just move the old delegation to the current
2955 	     * delegation list and issue open. If not found, return
2956 	     * NFSERR_EXPIRED.
2957 	     */
2958 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2959 		if (stp->ls_lfp == lfp) {
2960 		    /* Found it */
2961 		    if (stp->ls_clp != clp)
2962 			panic("olddeleg clp");
2963 		    LIST_REMOVE(stp, ls_list);
2964 		    LIST_REMOVE(stp, ls_hash);
2965 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2966 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2967 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2968 			clp->lc_clientid.lval[0];
2969 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2970 			clp->lc_clientid.lval[1];
2971 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2972 			nfsrv_nextstateindex(clp);
2973 		    stp->ls_compref = nd->nd_compref;
2974 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2975 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2976 			stp->ls_stateid), stp, ls_hash);
2977 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2978 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2979 		    else
2980 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2981 		    clp->lc_delegtime = NFSD_MONOSEC +
2982 			nfsrv_lease + NFSRV_LEASEDELTA;
2983 
2984 		    /*
2985 		     * Now, do the associated open.
2986 		     */
2987 		    new_open->ls_stateid.seqid = 1;
2988 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2989 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2990 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2991 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2992 			NFSLCK_OPEN;
2993 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2994 			new_open->ls_flags |= (NFSLCK_READACCESS |
2995 			    NFSLCK_WRITEACCESS);
2996 		    else
2997 			new_open->ls_flags |= NFSLCK_READACCESS;
2998 		    new_open->ls_uid = new_stp->ls_uid;
2999 		    new_open->ls_lfp = lfp;
3000 		    new_open->ls_clp = clp;
3001 		    LIST_INIT(&new_open->ls_open);
3002 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3003 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3004 			new_open, ls_hash);
3005 		    /*
3006 		     * and handle the open owner
3007 		     */
3008 		    if (ownerstp) {
3009 			new_open->ls_openowner = ownerstp;
3010 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
3011 		    } else {
3012 			new_open->ls_openowner = new_stp;
3013 			new_stp->ls_flags = 0;
3014 			nfsrvd_refcache(new_stp->ls_op);
3015 			new_stp->ls_noopens = 0;
3016 			LIST_INIT(&new_stp->ls_open);
3017 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3018 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3019 			*new_stpp = NULL;
3020 			nfsstatsv1.srvopenowners++;
3021 			nfsrv_openpluslock++;
3022 		    }
3023 		    openstp = new_open;
3024 		    new_open = NULL;
3025 		    nfsstatsv1.srvopens++;
3026 		    nfsrv_openpluslock++;
3027 		    break;
3028 		}
3029 	    }
3030 	    if (stp == LIST_END(&clp->lc_olddeleg))
3031 		error = NFSERR_EXPIRED;
3032 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
3033 	    /*
3034 	     * Scan to see that no delegation for this client and file
3035 	     * doesn't already exist.
3036 	     * There also shouldn't yet be an Open for this file and
3037 	     * openowner.
3038 	     */
3039 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
3040 		if (stp->ls_clp == clp)
3041 		    break;
3042 	    }
3043 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
3044 		/*
3045 		 * This is the Claim_Previous case with a delegation
3046 		 * type != Delegate_None.
3047 		 */
3048 		/*
3049 		 * First, add the delegation. (Although we must issue the
3050 		 * delegation, we can also ask for an immediate return.)
3051 		 */
3052 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3053 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
3054 		    clp->lc_clientid.lval[0];
3055 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
3056 		    clp->lc_clientid.lval[1];
3057 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
3058 		    nfsrv_nextstateindex(clp);
3059 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
3060 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3061 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3062 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3063 		    nfsrv_writedelegcnt++;
3064 		} else {
3065 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3066 			NFSLCK_READACCESS);
3067 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
3068 		}
3069 		new_deleg->ls_uid = new_stp->ls_uid;
3070 		new_deleg->ls_lfp = lfp;
3071 		new_deleg->ls_clp = clp;
3072 		new_deleg->ls_filerev = filerev;
3073 		new_deleg->ls_compref = nd->nd_compref;
3074 		new_deleg->ls_lastrecall = 0;
3075 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3076 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3077 		    new_deleg->ls_stateid), new_deleg, ls_hash);
3078 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3079 		new_deleg = NULL;
3080 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3081 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3082 		     LCL_CALLBACKSON ||
3083 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3084 		    !NFSVNO_DELEGOK(vp))
3085 		    *rflagsp |= NFSV4OPEN_RECALL;
3086 		nfsstatsv1.srvdelegates++;
3087 		nfsrv_openpluslock++;
3088 		nfsrv_delegatecnt++;
3089 
3090 		/*
3091 		 * Now, do the associated open.
3092 		 */
3093 		new_open->ls_stateid.seqid = 1;
3094 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3095 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3096 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3097 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3098 		    NFSLCK_OPEN;
3099 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3100 			new_open->ls_flags |= (NFSLCK_READACCESS |
3101 			    NFSLCK_WRITEACCESS);
3102 		else
3103 			new_open->ls_flags |= NFSLCK_READACCESS;
3104 		new_open->ls_uid = new_stp->ls_uid;
3105 		new_open->ls_lfp = lfp;
3106 		new_open->ls_clp = clp;
3107 		LIST_INIT(&new_open->ls_open);
3108 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3109 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3110 		   new_open, ls_hash);
3111 		/*
3112 		 * and handle the open owner
3113 		 */
3114 		if (ownerstp) {
3115 		    new_open->ls_openowner = ownerstp;
3116 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3117 		} else {
3118 		    new_open->ls_openowner = new_stp;
3119 		    new_stp->ls_flags = 0;
3120 		    nfsrvd_refcache(new_stp->ls_op);
3121 		    new_stp->ls_noopens = 0;
3122 		    LIST_INIT(&new_stp->ls_open);
3123 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3124 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3125 		    *new_stpp = NULL;
3126 		    nfsstatsv1.srvopenowners++;
3127 		    nfsrv_openpluslock++;
3128 		}
3129 		openstp = new_open;
3130 		new_open = NULL;
3131 		nfsstatsv1.srvopens++;
3132 		nfsrv_openpluslock++;
3133 	    } else {
3134 		error = NFSERR_RECLAIMCONFLICT;
3135 	    }
3136 	} else if (ownerstp) {
3137 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3138 		    /* Replace the open */
3139 		    if (ownerstp->ls_op)
3140 			nfsrvd_derefcache(ownerstp->ls_op);
3141 		    ownerstp->ls_op = new_stp->ls_op;
3142 		    nfsrvd_refcache(ownerstp->ls_op);
3143 		    ownerstp->ls_seq = new_stp->ls_seq;
3144 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3145 		    stp = LIST_FIRST(&ownerstp->ls_open);
3146 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3147 			NFSLCK_OPEN;
3148 		    stp->ls_stateid.seqid = 1;
3149 		    stp->ls_uid = new_stp->ls_uid;
3150 		    if (lfp != stp->ls_lfp) {
3151 			LIST_REMOVE(stp, ls_file);
3152 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3153 			stp->ls_lfp = lfp;
3154 		    }
3155 		    openstp = stp;
3156 		} else if (openstp) {
3157 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3158 		    openstp->ls_stateid.seqid++;
3159 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3160 			openstp->ls_stateid.seqid == 0)
3161 			openstp->ls_stateid.seqid = 1;
3162 
3163 		    /*
3164 		     * This is where we can choose to issue a delegation.
3165 		     */
3166 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3167 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3168 		    else if (nfsrv_issuedelegs == 0)
3169 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3170 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3171 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3172 		    else if (delegate == 0 || writedeleg == 0 ||
3173 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3174 			nfsrv_writedelegifpos == 0) ||
3175 			!NFSVNO_DELEGOK(vp) ||
3176 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3177 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3178 			 LCL_CALLBACKSON)
3179 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3180 		    else {
3181 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3182 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3183 			    = clp->lc_clientid.lval[0];
3184 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3185 			    = clp->lc_clientid.lval[1];
3186 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3187 			    = nfsrv_nextstateindex(clp);
3188 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3189 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3190 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3191 			new_deleg->ls_uid = new_stp->ls_uid;
3192 			new_deleg->ls_lfp = lfp;
3193 			new_deleg->ls_clp = clp;
3194 			new_deleg->ls_filerev = filerev;
3195 			new_deleg->ls_compref = nd->nd_compref;
3196 			new_deleg->ls_lastrecall = 0;
3197 			nfsrv_writedelegcnt++;
3198 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3199 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3200 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3201 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3202 			new_deleg = NULL;
3203 			nfsstatsv1.srvdelegates++;
3204 			nfsrv_openpluslock++;
3205 			nfsrv_delegatecnt++;
3206 		    }
3207 		} else {
3208 		    new_open->ls_stateid.seqid = 1;
3209 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3210 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3211 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3212 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3213 			NFSLCK_OPEN;
3214 		    new_open->ls_uid = new_stp->ls_uid;
3215 		    new_open->ls_openowner = ownerstp;
3216 		    new_open->ls_lfp = lfp;
3217 		    new_open->ls_clp = clp;
3218 		    LIST_INIT(&new_open->ls_open);
3219 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3220 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3221 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3222 			new_open, ls_hash);
3223 		    openstp = new_open;
3224 		    new_open = NULL;
3225 		    nfsstatsv1.srvopens++;
3226 		    nfsrv_openpluslock++;
3227 
3228 		    /*
3229 		     * This is where we can choose to issue a delegation.
3230 		     */
3231 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3232 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3233 		    else if (nfsrv_issuedelegs == 0)
3234 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3235 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3236 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3237 		    else if (delegate == 0 || (writedeleg == 0 &&
3238 			readonly == 0) || !NFSVNO_DELEGOK(vp) ||
3239 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3240 			 LCL_CALLBACKSON)
3241 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3242 		    else {
3243 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3244 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3245 			    = clp->lc_clientid.lval[0];
3246 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3247 			    = clp->lc_clientid.lval[1];
3248 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3249 			    = nfsrv_nextstateindex(clp);
3250 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3251 			    (nfsrv_writedelegifpos || !readonly) &&
3252 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3253 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3254 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3255 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3256 			    nfsrv_writedelegcnt++;
3257 			} else {
3258 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3259 				NFSLCK_READACCESS);
3260 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3261 			}
3262 			new_deleg->ls_uid = new_stp->ls_uid;
3263 			new_deleg->ls_lfp = lfp;
3264 			new_deleg->ls_clp = clp;
3265 			new_deleg->ls_filerev = filerev;
3266 			new_deleg->ls_compref = nd->nd_compref;
3267 			new_deleg->ls_lastrecall = 0;
3268 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3269 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3270 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3271 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3272 			new_deleg = NULL;
3273 			nfsstatsv1.srvdelegates++;
3274 			nfsrv_openpluslock++;
3275 			nfsrv_delegatecnt++;
3276 		    }
3277 		}
3278 	} else {
3279 		/*
3280 		 * New owner case. Start the open_owner sequence with a
3281 		 * Needs confirmation (unless a reclaim) and hang the
3282 		 * new open off it.
3283 		 */
3284 		new_open->ls_stateid.seqid = 1;
3285 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3286 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3287 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3288 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3289 		    NFSLCK_OPEN;
3290 		new_open->ls_uid = new_stp->ls_uid;
3291 		LIST_INIT(&new_open->ls_open);
3292 		new_open->ls_openowner = new_stp;
3293 		new_open->ls_lfp = lfp;
3294 		new_open->ls_clp = clp;
3295 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3296 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3297 			new_stp->ls_flags = 0;
3298 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3299 			/* NFSv4.1 never needs confirmation. */
3300 			new_stp->ls_flags = 0;
3301 
3302 			/*
3303 			 * This is where we can choose to issue a delegation.
3304 			 */
3305 			if (delegate && nfsrv_issuedelegs &&
3306 			    (writedeleg || readonly) &&
3307 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3308 			     LCL_CALLBACKSON &&
3309 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3310 			    NFSVNO_DELEGOK(vp) &&
3311 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3312 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3313 				new_deleg->ls_stateid.seqid =
3314 				    delegstateidp->seqid = 1;
3315 				new_deleg->ls_stateid.other[0] =
3316 				    delegstateidp->other[0]
3317 				    = clp->lc_clientid.lval[0];
3318 				new_deleg->ls_stateid.other[1] =
3319 				    delegstateidp->other[1]
3320 				    = clp->lc_clientid.lval[1];
3321 				new_deleg->ls_stateid.other[2] =
3322 				    delegstateidp->other[2]
3323 				    = nfsrv_nextstateindex(clp);
3324 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3325 				    (nfsrv_writedelegifpos || !readonly) &&
3326 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3327 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3328 				     0)) {
3329 					new_deleg->ls_flags =
3330 					    (NFSLCK_DELEGWRITE |
3331 					     NFSLCK_READACCESS |
3332 					     NFSLCK_WRITEACCESS);
3333 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3334 					nfsrv_writedelegcnt++;
3335 				} else {
3336 					new_deleg->ls_flags =
3337 					    (NFSLCK_DELEGREAD |
3338 					     NFSLCK_READACCESS);
3339 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3340 				}
3341 				new_deleg->ls_uid = new_stp->ls_uid;
3342 				new_deleg->ls_lfp = lfp;
3343 				new_deleg->ls_clp = clp;
3344 				new_deleg->ls_filerev = filerev;
3345 				new_deleg->ls_compref = nd->nd_compref;
3346 				new_deleg->ls_lastrecall = 0;
3347 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3348 				    ls_file);
3349 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3350 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3351 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3352 				    ls_list);
3353 				new_deleg = NULL;
3354 				nfsstatsv1.srvdelegates++;
3355 				nfsrv_openpluslock++;
3356 				nfsrv_delegatecnt++;
3357 			}
3358 			/*
3359 			 * Since NFSv4.1 never does an OpenConfirm, the first
3360 			 * open state will be acquired here.
3361 			 */
3362 			if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3363 				clp->lc_flags |= LCL_STAMPEDSTABLE;
3364 				len = clp->lc_idlen;
3365 				NFSBCOPY(clp->lc_id, clidp, len);
3366 				gotstate = 1;
3367 			}
3368 		} else {
3369 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3370 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3371 		}
3372 		nfsrvd_refcache(new_stp->ls_op);
3373 		new_stp->ls_noopens = 0;
3374 		LIST_INIT(&new_stp->ls_open);
3375 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3376 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3377 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3378 		    new_open, ls_hash);
3379 		openstp = new_open;
3380 		new_open = NULL;
3381 		*new_stpp = NULL;
3382 		nfsstatsv1.srvopens++;
3383 		nfsrv_openpluslock++;
3384 		nfsstatsv1.srvopenowners++;
3385 		nfsrv_openpluslock++;
3386 	}
3387 	if (!error) {
3388 		stateidp->seqid = openstp->ls_stateid.seqid;
3389 		stateidp->other[0] = openstp->ls_stateid.other[0];
3390 		stateidp->other[1] = openstp->ls_stateid.other[1];
3391 		stateidp->other[2] = openstp->ls_stateid.other[2];
3392 	}
3393 	NFSUNLOCKSTATE();
3394 	if (haslock) {
3395 		NFSLOCKV4ROOTMUTEX();
3396 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3397 		NFSUNLOCKV4ROOTMUTEX();
3398 	}
3399 	if (new_open)
3400 		free(new_open, M_NFSDSTATE);
3401 	if (new_deleg)
3402 		free(new_deleg, M_NFSDSTATE);
3403 
3404 	/*
3405 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3406 	 * to the stable storage file.
3407 	 */
3408 	if (gotstate != 0) {
3409 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3410 		nfsrv_backupstable();
3411 	}
3412 
3413 out:
3414 	free(clidp, M_TEMP);
3415 	NFSEXITCODE2(error, nd);
3416 	return (error);
3417 }
3418 
3419 /*
3420  * Open update. Does the confirm, downgrade and close.
3421  */
3422 int
3423 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3424     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p,
3425     int *retwriteaccessp)
3426 {
3427 	struct nfsstate *stp;
3428 	struct nfsclient *clp;
3429 	struct nfslockfile *lfp;
3430 	u_int32_t bits;
3431 	int error = 0, gotstate = 0, len = 0;
3432 	u_char *clidp = NULL;
3433 
3434 	/*
3435 	 * Check for restart conditions (client and server).
3436 	 */
3437 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3438 	    &new_stp->ls_stateid, 0);
3439 	if (error)
3440 		goto out;
3441 
3442 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3443 	NFSLOCKSTATE();
3444 	/*
3445 	 * Get the open structure via clientid and stateid.
3446 	 */
3447 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3448 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3449 	if (!error)
3450 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3451 		    new_stp->ls_flags, &stp);
3452 
3453 	/*
3454 	 * Sanity check the open.
3455 	 */
3456 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3457 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3458 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3459 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3460 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3461 		error = NFSERR_BADSTATEID;
3462 
3463 	if (!error)
3464 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3465 		    stp->ls_openowner, new_stp->ls_op);
3466 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3467 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3468 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3469 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3470 	      new_stp->ls_stateid.seqid != 0)))
3471 		error = NFSERR_OLDSTATEID;
3472 	if (!error && vnode_vtype(vp) != VREG) {
3473 		if (vnode_vtype(vp) == VDIR)
3474 			error = NFSERR_ISDIR;
3475 		else
3476 			error = NFSERR_INVAL;
3477 	}
3478 
3479 	if (error) {
3480 		/*
3481 		 * If a client tries to confirm an Open with a bad
3482 		 * seqid# and there are no byte range locks or other Opens
3483 		 * on the openowner, just throw it away, so the next use of the
3484 		 * openowner will start a fresh seq#.
3485 		 */
3486 		if (error == NFSERR_BADSEQID &&
3487 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3488 		    nfsrv_nootherstate(stp))
3489 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3490 		NFSUNLOCKSTATE();
3491 		goto out;
3492 	}
3493 
3494 	/*
3495 	 * Set the return stateid.
3496 	 */
3497 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3498 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3499 		stateidp->seqid = 1;
3500 	stateidp->other[0] = stp->ls_stateid.other[0];
3501 	stateidp->other[1] = stp->ls_stateid.other[1];
3502 	stateidp->other[2] = stp->ls_stateid.other[2];
3503 	/*
3504 	 * Now, handle the three cases.
3505 	 */
3506 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3507 		/*
3508 		 * If the open doesn't need confirmation, it seems to me that
3509 		 * there is a client error, but I'll just log it and keep going?
3510 		 */
3511 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3512 			printf("Nfsv4d: stray open confirm\n");
3513 		stp->ls_openowner->ls_flags = 0;
3514 		stp->ls_stateid.seqid++;
3515 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3516 		    stp->ls_stateid.seqid == 0)
3517 			stp->ls_stateid.seqid = 1;
3518 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3519 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3520 			len = clp->lc_idlen;
3521 			NFSBCOPY(clp->lc_id, clidp, len);
3522 			gotstate = 1;
3523 		}
3524 		NFSUNLOCKSTATE();
3525 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3526 		lfp = stp->ls_lfp;
3527 		if (retwriteaccessp != NULL) {
3528 			if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0)
3529 				*retwriteaccessp = 1;
3530 			else
3531 				*retwriteaccessp = 0;
3532 		}
3533 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3534 			/* Get the lf lock */
3535 			nfsrv_locklf(lfp);
3536 			NFSUNLOCKSTATE();
3537 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3538 			NFSVOPUNLOCK(vp);
3539 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3540 				NFSLOCKSTATE();
3541 				nfsrv_unlocklf(lfp);
3542 				NFSUNLOCKSTATE();
3543 			}
3544 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3545 		} else {
3546 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3547 			NFSUNLOCKSTATE();
3548 		}
3549 	} else {
3550 		/*
3551 		 * Update the share bits, making sure that the new set are a
3552 		 * subset of the old ones.
3553 		 */
3554 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3555 		if (~(stp->ls_flags) & bits) {
3556 			NFSUNLOCKSTATE();
3557 			error = NFSERR_INVAL;
3558 			goto out;
3559 		}
3560 		stp->ls_flags = (bits | NFSLCK_OPEN);
3561 		stp->ls_stateid.seqid++;
3562 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3563 		    stp->ls_stateid.seqid == 0)
3564 			stp->ls_stateid.seqid = 1;
3565 		NFSUNLOCKSTATE();
3566 	}
3567 
3568 	/*
3569 	 * If the client just confirmed its first open, write a timestamp
3570 	 * to the stable storage file.
3571 	 */
3572 	if (gotstate != 0) {
3573 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3574 		nfsrv_backupstable();
3575 	}
3576 
3577 out:
3578 	free(clidp, M_TEMP);
3579 	NFSEXITCODE2(error, nd);
3580 	return (error);
3581 }
3582 
3583 /*
3584  * Delegation update. Does the purge and return.
3585  */
3586 int
3587 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3588     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3589     NFSPROC_T *p, int *retwriteaccessp)
3590 {
3591 	struct nfsstate *stp;
3592 	struct nfsclient *clp;
3593 	int error = 0;
3594 	fhandle_t fh;
3595 
3596 	/*
3597 	 * Do a sanity check against the file handle for DelegReturn.
3598 	 */
3599 	if (vp) {
3600 		error = nfsvno_getfh(vp, &fh, p);
3601 		if (error)
3602 			goto out;
3603 	}
3604 	/*
3605 	 * Check for restart conditions (client and server).
3606 	 */
3607 	if (op == NFSV4OP_DELEGRETURN)
3608 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3609 			stateidp, 0);
3610 	else
3611 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3612 			stateidp, 0);
3613 
3614 	NFSLOCKSTATE();
3615 	/*
3616 	 * Get the open structure via clientid and stateid.
3617 	 */
3618 	if (!error)
3619 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3620 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3621 	if (error) {
3622 		if (error == NFSERR_CBPATHDOWN)
3623 			error = 0;
3624 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3625 			error = NFSERR_STALESTATEID;
3626 	}
3627 	if (!error && op == NFSV4OP_DELEGRETURN) {
3628 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3629 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3630 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3631 		error = NFSERR_OLDSTATEID;
3632 	}
3633 	/*
3634 	 * NFSERR_EXPIRED means that the state has gone away,
3635 	 * so Delegations have been purged. Just return ok.
3636 	 */
3637 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3638 		NFSUNLOCKSTATE();
3639 		error = 0;
3640 		goto out;
3641 	}
3642 	if (error) {
3643 		NFSUNLOCKSTATE();
3644 		goto out;
3645 	}
3646 
3647 	if (op == NFSV4OP_DELEGRETURN) {
3648 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3649 		    sizeof (fhandle_t))) {
3650 			NFSUNLOCKSTATE();
3651 			error = NFSERR_BADSTATEID;
3652 			goto out;
3653 		}
3654 		if (retwriteaccessp != NULL) {
3655 			if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
3656 				*retwriteaccessp = 1;
3657 			else
3658 				*retwriteaccessp = 0;
3659 		}
3660 		nfsrv_freedeleg(stp);
3661 	} else {
3662 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3663 	}
3664 	NFSUNLOCKSTATE();
3665 	error = 0;
3666 
3667 out:
3668 	NFSEXITCODE(error);
3669 	return (error);
3670 }
3671 
3672 /*
3673  * Release lock owner.
3674  */
3675 int
3676 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3677     NFSPROC_T *p)
3678 {
3679 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3680 	struct nfsclient *clp;
3681 	int error = 0;
3682 
3683 	/*
3684 	 * Check for restart conditions (client and server).
3685 	 */
3686 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3687 	    &new_stp->ls_stateid, 0);
3688 	if (error)
3689 		goto out;
3690 
3691 	NFSLOCKSTATE();
3692 	/*
3693 	 * Get the lock owner by name.
3694 	 */
3695 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3696 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3697 	if (error) {
3698 		NFSUNLOCKSTATE();
3699 		goto out;
3700 	}
3701 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3702 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3703 		stp = LIST_FIRST(&openstp->ls_open);
3704 		while (stp != LIST_END(&openstp->ls_open)) {
3705 		    nstp = LIST_NEXT(stp, ls_list);
3706 		    /*
3707 		     * If the owner matches, check for locks and
3708 		     * then free or return an error.
3709 		     */
3710 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3711 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3712 			 stp->ls_ownerlen)){
3713 			if (LIST_EMPTY(&stp->ls_lock)) {
3714 			    nfsrv_freelockowner(stp, NULL, 0, p);
3715 			} else {
3716 			    NFSUNLOCKSTATE();
3717 			    error = NFSERR_LOCKSHELD;
3718 			    goto out;
3719 			}
3720 		    }
3721 		    stp = nstp;
3722 		}
3723 	    }
3724 	}
3725 	NFSUNLOCKSTATE();
3726 
3727 out:
3728 	NFSEXITCODE(error);
3729 	return (error);
3730 }
3731 
3732 /*
3733  * Get the file handle for a lock structure.
3734  */
3735 static int
3736 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3737     fhandle_t *nfhp, NFSPROC_T *p)
3738 {
3739 	fhandle_t *fhp = NULL;
3740 	int error;
3741 
3742 	/*
3743 	 * For lock, use the new nfslock structure, otherwise just
3744 	 * a fhandle_t on the stack.
3745 	 */
3746 	if (flags & NFSLCK_OPEN) {
3747 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3748 		fhp = &new_lfp->lf_fh;
3749 	} else if (nfhp) {
3750 		fhp = nfhp;
3751 	} else {
3752 		panic("nfsrv_getlockfh");
3753 	}
3754 	error = nfsvno_getfh(vp, fhp, p);
3755 	NFSEXITCODE(error);
3756 	return (error);
3757 }
3758 
3759 /*
3760  * Get an nfs lock structure. Allocate one, as required, and return a
3761  * pointer to it.
3762  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3763  */
3764 static int
3765 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3766     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3767 {
3768 	struct nfslockfile *lfp;
3769 	fhandle_t *fhp = NULL, *tfhp;
3770 	struct nfslockhashhead *hp;
3771 	struct nfslockfile *new_lfp = NULL;
3772 
3773 	/*
3774 	 * For lock, use the new nfslock structure, otherwise just
3775 	 * a fhandle_t on the stack.
3776 	 */
3777 	if (flags & NFSLCK_OPEN) {
3778 		new_lfp = *new_lfpp;
3779 		fhp = &new_lfp->lf_fh;
3780 	} else if (nfhp) {
3781 		fhp = nfhp;
3782 	} else {
3783 		panic("nfsrv_getlockfile");
3784 	}
3785 
3786 	hp = NFSLOCKHASH(fhp);
3787 	LIST_FOREACH(lfp, hp, lf_hash) {
3788 		tfhp = &lfp->lf_fh;
3789 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3790 			if (lockit)
3791 				nfsrv_locklf(lfp);
3792 			*lfpp = lfp;
3793 			return (0);
3794 		}
3795 	}
3796 	if (!(flags & NFSLCK_OPEN))
3797 		return (-1);
3798 
3799 	/*
3800 	 * No match, so chain the new one into the list.
3801 	 */
3802 	LIST_INIT(&new_lfp->lf_open);
3803 	LIST_INIT(&new_lfp->lf_lock);
3804 	LIST_INIT(&new_lfp->lf_deleg);
3805 	LIST_INIT(&new_lfp->lf_locallock);
3806 	LIST_INIT(&new_lfp->lf_rollback);
3807 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3808 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3809 	new_lfp->lf_usecount = 0;
3810 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3811 	*lfpp = new_lfp;
3812 	*new_lfpp = NULL;
3813 	return (0);
3814 }
3815 
3816 /*
3817  * This function adds a nfslock lock structure to the list for the associated
3818  * nfsstate and nfslockfile structures. It will be inserted after the
3819  * entry pointed at by insert_lop.
3820  */
3821 static void
3822 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3823     struct nfsstate *stp, struct nfslockfile *lfp)
3824 {
3825 	struct nfslock *lop, *nlop;
3826 
3827 	new_lop->lo_stp = stp;
3828 	new_lop->lo_lfp = lfp;
3829 
3830 	if (stp != NULL) {
3831 		/* Insert in increasing lo_first order */
3832 		lop = LIST_FIRST(&lfp->lf_lock);
3833 		if (lop == LIST_END(&lfp->lf_lock) ||
3834 		    new_lop->lo_first <= lop->lo_first) {
3835 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3836 		} else {
3837 			nlop = LIST_NEXT(lop, lo_lckfile);
3838 			while (nlop != LIST_END(&lfp->lf_lock) &&
3839 			       nlop->lo_first < new_lop->lo_first) {
3840 				lop = nlop;
3841 				nlop = LIST_NEXT(lop, lo_lckfile);
3842 			}
3843 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3844 		}
3845 	} else {
3846 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3847 	}
3848 
3849 	/*
3850 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3851 	 * an empty list.
3852 	 */
3853 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3854 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3855 	else if ((struct nfsstate *)insert_lop == stp)
3856 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3857 	else
3858 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3859 	if (stp != NULL) {
3860 		nfsstatsv1.srvlocks++;
3861 		nfsrv_openpluslock++;
3862 	}
3863 }
3864 
3865 /*
3866  * This function updates the locking for a lock owner and given file. It
3867  * maintains a list of lock ranges ordered on increasing file offset that
3868  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3869  * It always adds new_lop to the list and sometimes uses the one pointed
3870  * at by other_lopp.
3871  */
3872 static void
3873 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3874     struct nfslock **other_lopp, struct nfslockfile *lfp)
3875 {
3876 	struct nfslock *new_lop = *new_lopp;
3877 	struct nfslock *lop, *tlop, *ilop;
3878 	struct nfslock *other_lop = *other_lopp;
3879 	int unlock = 0, myfile = 0;
3880 	u_int64_t tmp;
3881 
3882 	/*
3883 	 * Work down the list until the lock is merged.
3884 	 */
3885 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3886 		unlock = 1;
3887 	if (stp != NULL) {
3888 		ilop = (struct nfslock *)stp;
3889 		lop = LIST_FIRST(&stp->ls_lock);
3890 	} else {
3891 		ilop = (struct nfslock *)lfp;
3892 		lop = LIST_FIRST(&lfp->lf_locallock);
3893 	}
3894 	while (lop != NULL) {
3895 	    /*
3896 	     * Only check locks for this file that aren't before the start of
3897 	     * new lock's range.
3898 	     */
3899 	    if (lop->lo_lfp == lfp) {
3900 	      myfile = 1;
3901 	      if (lop->lo_end >= new_lop->lo_first) {
3902 		if (new_lop->lo_end < lop->lo_first) {
3903 			/*
3904 			 * If the new lock ends before the start of the
3905 			 * current lock's range, no merge, just insert
3906 			 * the new lock.
3907 			 */
3908 			break;
3909 		}
3910 		if (new_lop->lo_flags == lop->lo_flags ||
3911 		    (new_lop->lo_first <= lop->lo_first &&
3912 		     new_lop->lo_end >= lop->lo_end)) {
3913 			/*
3914 			 * This lock can be absorbed by the new lock/unlock.
3915 			 * This happens when it covers the entire range
3916 			 * of the old lock or is contiguous
3917 			 * with the old lock and is of the same type or an
3918 			 * unlock.
3919 			 */
3920 			if (lop->lo_first < new_lop->lo_first)
3921 				new_lop->lo_first = lop->lo_first;
3922 			if (lop->lo_end > new_lop->lo_end)
3923 				new_lop->lo_end = lop->lo_end;
3924 			tlop = lop;
3925 			lop = LIST_NEXT(lop, lo_lckowner);
3926 			nfsrv_freenfslock(tlop);
3927 			continue;
3928 		}
3929 
3930 		/*
3931 		 * All these cases are for contiguous locks that are not the
3932 		 * same type, so they can't be merged.
3933 		 */
3934 		if (new_lop->lo_first <= lop->lo_first) {
3935 			/*
3936 			 * This case is where the new lock overlaps with the
3937 			 * first part of the old lock. Move the start of the
3938 			 * old lock to just past the end of the new lock. The
3939 			 * new lock will be inserted in front of the old, since
3940 			 * ilop hasn't been updated. (We are done now.)
3941 			 */
3942 			lop->lo_first = new_lop->lo_end;
3943 			break;
3944 		}
3945 		if (new_lop->lo_end >= lop->lo_end) {
3946 			/*
3947 			 * This case is where the new lock overlaps with the
3948 			 * end of the old lock's range. Move the old lock's
3949 			 * end to just before the new lock's first and insert
3950 			 * the new lock after the old lock.
3951 			 * Might not be done yet, since the new lock could
3952 			 * overlap further locks with higher ranges.
3953 			 */
3954 			lop->lo_end = new_lop->lo_first;
3955 			ilop = lop;
3956 			lop = LIST_NEXT(lop, lo_lckowner);
3957 			continue;
3958 		}
3959 		/*
3960 		 * The final case is where the new lock's range is in the
3961 		 * middle of the current lock's and splits the current lock
3962 		 * up. Use *other_lopp to handle the second part of the
3963 		 * split old lock range. (We are done now.)
3964 		 * For unlock, we use new_lop as other_lop and tmp, since
3965 		 * other_lop and new_lop are the same for this case.
3966 		 * We noted the unlock case above, so we don't need
3967 		 * new_lop->lo_flags any longer.
3968 		 */
3969 		tmp = new_lop->lo_first;
3970 		if (other_lop == NULL) {
3971 			if (!unlock)
3972 				panic("nfsd srv update unlock");
3973 			other_lop = new_lop;
3974 			*new_lopp = NULL;
3975 		}
3976 		other_lop->lo_first = new_lop->lo_end;
3977 		other_lop->lo_end = lop->lo_end;
3978 		other_lop->lo_flags = lop->lo_flags;
3979 		other_lop->lo_stp = stp;
3980 		other_lop->lo_lfp = lfp;
3981 		lop->lo_end = tmp;
3982 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3983 		*other_lopp = NULL;
3984 		ilop = lop;
3985 		break;
3986 	      }
3987 	    }
3988 	    ilop = lop;
3989 	    lop = LIST_NEXT(lop, lo_lckowner);
3990 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3991 		break;
3992 	}
3993 
3994 	/*
3995 	 * Insert the new lock in the list at the appropriate place.
3996 	 */
3997 	if (!unlock) {
3998 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3999 		*new_lopp = NULL;
4000 	}
4001 }
4002 
4003 /*
4004  * This function handles sequencing of locks, etc.
4005  * It returns an error that indicates what the caller should do.
4006  */
4007 static int
4008 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
4009     struct nfsstate *stp, struct nfsrvcache *op)
4010 {
4011 	int error = 0;
4012 
4013 	if ((nd->nd_flag & ND_NFSV41) != 0)
4014 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
4015 		goto out;
4016 	if (op != nd->nd_rp)
4017 		panic("nfsrvstate checkseqid");
4018 	if (!(op->rc_flag & RC_INPROG))
4019 		panic("nfsrvstate not inprog");
4020 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
4021 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
4022 		panic("nfsrvstate op refcnt");
4023 	}
4024 
4025 	/* If ND_ERELOOKUP is set, the seqid has already been handled. */
4026 	if ((nd->nd_flag & ND_ERELOOKUP) != 0)
4027 		goto out;
4028 
4029 	if ((stp->ls_seq + 1) == seqid) {
4030 		if (stp->ls_op)
4031 			nfsrvd_derefcache(stp->ls_op);
4032 		stp->ls_op = op;
4033 		nfsrvd_refcache(op);
4034 		stp->ls_seq = seqid;
4035 		goto out;
4036 	} else if (stp->ls_seq == seqid && stp->ls_op &&
4037 		op->rc_xid == stp->ls_op->rc_xid &&
4038 		op->rc_refcnt == 0 &&
4039 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
4040 		op->rc_cksum == stp->ls_op->rc_cksum) {
4041 		if (stp->ls_op->rc_flag & RC_INPROG) {
4042 			error = NFSERR_DONTREPLY;
4043 			goto out;
4044 		}
4045 		nd->nd_rp = stp->ls_op;
4046 		nd->nd_rp->rc_flag |= RC_INPROG;
4047 		nfsrvd_delcache(op);
4048 		error = NFSERR_REPLYFROMCACHE;
4049 		goto out;
4050 	}
4051 	error = NFSERR_BADSEQID;
4052 
4053 out:
4054 	NFSEXITCODE2(error, nd);
4055 	return (error);
4056 }
4057 
4058 /*
4059  * Get the client ip address for callbacks. If the strings can't be parsed,
4060  * just set lc_program to 0 to indicate no callbacks are possible.
4061  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
4062  *  the address to the client's transport address. This won't be used
4063  *  for callbacks, but can be printed out by nfsstats for info.)
4064  * Return error if the xdr can't be parsed, 0 otherwise.
4065  */
4066 int
4067 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
4068 {
4069 	u_int32_t *tl;
4070 	u_char *cp, *cp2;
4071 	int i, j, maxalen = 0, minalen = 0;
4072 	sa_family_t af;
4073 #ifdef INET
4074 	struct sockaddr_in *rin = NULL, *sin;
4075 #endif
4076 #ifdef INET6
4077 	struct sockaddr_in6 *rin6 = NULL, *sin6;
4078 #endif
4079 	u_char *addr;
4080 	int error = 0, cantparse = 0;
4081 	union {
4082 		in_addr_t ival;
4083 		u_char cval[4];
4084 	} ip;
4085 	union {
4086 		in_port_t sval;
4087 		u_char cval[2];
4088 	} port;
4089 
4090 	/* 8 is the maximum length of the port# string. */
4091 	addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK);
4092 	clp->lc_req.nr_client = NULL;
4093 	clp->lc_req.nr_lock = 0;
4094 	af = AF_UNSPEC;
4095 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4096 	i = fxdr_unsigned(int, *tl);
4097 	if (i >= 3 && i <= 4) {
4098 		error = nfsrv_mtostr(nd, addr, i);
4099 		if (error)
4100 			goto nfsmout;
4101 #ifdef INET
4102 		if (!strcmp(addr, "tcp")) {
4103 			clp->lc_flags |= LCL_TCPCALLBACK;
4104 			clp->lc_req.nr_sotype = SOCK_STREAM;
4105 			clp->lc_req.nr_soproto = IPPROTO_TCP;
4106 			af = AF_INET;
4107 		} else if (!strcmp(addr, "udp")) {
4108 			clp->lc_req.nr_sotype = SOCK_DGRAM;
4109 			clp->lc_req.nr_soproto = IPPROTO_UDP;
4110 			af = AF_INET;
4111 		}
4112 #endif
4113 #ifdef INET6
4114 		if (af == AF_UNSPEC) {
4115 			if (!strcmp(addr, "tcp6")) {
4116 				clp->lc_flags |= LCL_TCPCALLBACK;
4117 				clp->lc_req.nr_sotype = SOCK_STREAM;
4118 				clp->lc_req.nr_soproto = IPPROTO_TCP;
4119 				af = AF_INET6;
4120 			} else if (!strcmp(addr, "udp6")) {
4121 				clp->lc_req.nr_sotype = SOCK_DGRAM;
4122 				clp->lc_req.nr_soproto = IPPROTO_UDP;
4123 				af = AF_INET6;
4124 			}
4125 		}
4126 #endif
4127 		if (af == AF_UNSPEC) {
4128 			cantparse = 1;
4129 		}
4130 	} else {
4131 		cantparse = 1;
4132 		if (i > 0) {
4133 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4134 			if (error)
4135 				goto nfsmout;
4136 		}
4137 	}
4138 	/*
4139 	 * The caller has allocated clp->lc_req.nr_nam to be large enough
4140 	 * for either AF_INET or AF_INET6 and zeroed out the contents.
4141 	 * maxalen is set to the maximum length of the host IP address string
4142 	 * plus 8 for the maximum length of the port#.
4143 	 * minalen is set to the minimum length of the host IP address string
4144 	 * plus 4 for the minimum length of the port#.
4145 	 * These lengths do not include NULL termination,
4146 	 * so INET[6]_ADDRSTRLEN - 1 is used in the calculations.
4147 	 */
4148 	switch (af) {
4149 #ifdef INET
4150 	case AF_INET:
4151 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4152 		rin->sin_family = AF_INET;
4153 		rin->sin_len = sizeof(struct sockaddr_in);
4154 		maxalen = INET_ADDRSTRLEN - 1 + 8;
4155 		minalen = 7 + 4;
4156 		break;
4157 #endif
4158 #ifdef INET6
4159 	case AF_INET6:
4160 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4161 		rin6->sin6_family = AF_INET6;
4162 		rin6->sin6_len = sizeof(struct sockaddr_in6);
4163 		maxalen = INET6_ADDRSTRLEN - 1 + 8;
4164 		minalen = 3 + 4;
4165 		break;
4166 #endif
4167 	}
4168 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4169 	i = fxdr_unsigned(int, *tl);
4170 	if (i < 0) {
4171 		error = NFSERR_BADXDR;
4172 		goto nfsmout;
4173 	} else if (i == 0) {
4174 		cantparse = 1;
4175 	} else if (!cantparse && i <= maxalen && i >= minalen) {
4176 		error = nfsrv_mtostr(nd, addr, i);
4177 		if (error)
4178 			goto nfsmout;
4179 
4180 		/*
4181 		 * Parse out the address fields. We expect 6 decimal numbers
4182 		 * separated by '.'s for AF_INET and two decimal numbers
4183 		 * preceeded by '.'s for AF_INET6.
4184 		 */
4185 		cp = NULL;
4186 		switch (af) {
4187 #ifdef INET6
4188 		/*
4189 		 * For AF_INET6, first parse the host address.
4190 		 */
4191 		case AF_INET6:
4192 			cp = strchr(addr, '.');
4193 			if (cp != NULL) {
4194 				*cp++ = '\0';
4195 				if (inet_pton(af, addr, &rin6->sin6_addr) == 1)
4196 					i = 4;
4197 				else {
4198 					cp = NULL;
4199 					cantparse = 1;
4200 				}
4201 			}
4202 			break;
4203 #endif
4204 #ifdef INET
4205 		case AF_INET:
4206 			cp = addr;
4207 			i = 0;
4208 			break;
4209 #endif
4210 		}
4211 		while (cp != NULL && *cp && i < 6) {
4212 			cp2 = cp;
4213 			while (*cp2 && *cp2 != '.')
4214 				cp2++;
4215 			if (*cp2)
4216 				*cp2++ = '\0';
4217 			else if (i != 5) {
4218 				cantparse = 1;
4219 				break;
4220 			}
4221 			j = nfsrv_getipnumber(cp);
4222 			if (j >= 0) {
4223 				if (i < 4)
4224 					ip.cval[3 - i] = j;
4225 				else
4226 					port.cval[5 - i] = j;
4227 			} else {
4228 				cantparse = 1;
4229 				break;
4230 			}
4231 			cp = cp2;
4232 			i++;
4233 		}
4234 		if (!cantparse) {
4235 			/*
4236 			 * The host address INADDR_ANY is (mis)used to indicate
4237 			 * "there is no valid callback address".
4238 			 */
4239 			switch (af) {
4240 #ifdef INET6
4241 			case AF_INET6:
4242 				if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr,
4243 				    &in6addr_any))
4244 					rin6->sin6_port = htons(port.sval);
4245 				else
4246 					cantparse = 1;
4247 				break;
4248 #endif
4249 #ifdef INET
4250 			case AF_INET:
4251 				if (ip.ival != INADDR_ANY) {
4252 					rin->sin_addr.s_addr = htonl(ip.ival);
4253 					rin->sin_port = htons(port.sval);
4254 				} else {
4255 					cantparse = 1;
4256 				}
4257 				break;
4258 #endif
4259 			}
4260 		}
4261 	} else {
4262 		cantparse = 1;
4263 		if (i > 0) {
4264 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4265 			if (error)
4266 				goto nfsmout;
4267 		}
4268 	}
4269 	if (cantparse) {
4270 		switch (nd->nd_nam->sa_family) {
4271 #ifdef INET
4272 		case AF_INET:
4273 			sin = (struct sockaddr_in *)nd->nd_nam;
4274 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4275 			rin->sin_family = AF_INET;
4276 			rin->sin_len = sizeof(struct sockaddr_in);
4277 			rin->sin_addr.s_addr = sin->sin_addr.s_addr;
4278 			rin->sin_port = 0x0;
4279 			break;
4280 #endif
4281 #ifdef INET6
4282 		case AF_INET6:
4283 			sin6 = (struct sockaddr_in6 *)nd->nd_nam;
4284 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4285 			rin6->sin6_family = AF_INET6;
4286 			rin6->sin6_len = sizeof(struct sockaddr_in6);
4287 			rin6->sin6_addr = sin6->sin6_addr;
4288 			rin6->sin6_port = 0x0;
4289 			break;
4290 #endif
4291 		}
4292 		clp->lc_program = 0;
4293 	}
4294 nfsmout:
4295 	free(addr, M_TEMP);
4296 	NFSEXITCODE2(error, nd);
4297 	return (error);
4298 }
4299 
4300 /*
4301  * Turn a string of up to three decimal digits into a number. Return -1 upon
4302  * error.
4303  */
4304 static int
4305 nfsrv_getipnumber(u_char *cp)
4306 {
4307 	int i = 0, j = 0;
4308 
4309 	while (*cp) {
4310 		if (j > 2 || *cp < '0' || *cp > '9')
4311 			return (-1);
4312 		i *= 10;
4313 		i += (*cp - '0');
4314 		cp++;
4315 		j++;
4316 	}
4317 	if (i < 256)
4318 		return (i);
4319 	return (-1);
4320 }
4321 
4322 /*
4323  * This function checks for restart conditions.
4324  */
4325 static int
4326 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4327     nfsv4stateid_t *stateidp, int specialid)
4328 {
4329 	int ret = 0;
4330 
4331 	/*
4332 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4333 	 * and DelegPurge have a clientid, the rest a stateid.
4334 	 */
4335 	if (flags &
4336 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4337 		if (clientid.lval[0] != nfsrvboottime) {
4338 			ret = NFSERR_STALECLIENTID;
4339 			goto out;
4340 		}
4341 	} else if (stateidp->other[0] != nfsrvboottime &&
4342 		specialid == 0) {
4343 		ret = NFSERR_STALESTATEID;
4344 		goto out;
4345 	}
4346 
4347 	/*
4348 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4349 	 * not use a lock/open owner seqid#, so the check can be done now.
4350 	 * (The others will be checked, as required, later.)
4351 	 */
4352 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4353 		goto out;
4354 
4355 	NFSLOCKSTATE();
4356 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4357 	NFSUNLOCKSTATE();
4358 
4359 out:
4360 	NFSEXITCODE(ret);
4361 	return (ret);
4362 }
4363 
4364 /*
4365  * Check for grace.
4366  */
4367 static int
4368 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4369     u_int32_t flags)
4370 {
4371 	int error = 0, notreclaimed;
4372 	struct nfsrv_stable *sp;
4373 
4374 	if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE |
4375 	     NFSNSF_GRACEOVER)) == 0) {
4376 		/*
4377 		 * First, check to see if all of the clients have done a
4378 		 * ReclaimComplete.  If so, grace can end now.
4379 		 */
4380 		notreclaimed = 0;
4381 		LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4382 			if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4383 				notreclaimed = 1;
4384 				break;
4385 			}
4386 		}
4387 		if (notreclaimed == 0)
4388 			nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER |
4389 			    NFSNSF_NEEDLOCK);
4390 	}
4391 
4392 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4393 		if (flags & NFSLCK_RECLAIM) {
4394 			error = NFSERR_NOGRACE;
4395 			goto out;
4396 		}
4397 	} else {
4398 		if (!(flags & NFSLCK_RECLAIM)) {
4399 			error = NFSERR_GRACE;
4400 			goto out;
4401 		}
4402 		if (nd != NULL && clp != NULL &&
4403 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4404 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4405 			error = NFSERR_NOGRACE;
4406 			goto out;
4407 		}
4408 
4409 		/*
4410 		 * If grace is almost over and we are still getting Reclaims,
4411 		 * extend grace a bit.
4412 		 */
4413 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4414 		    nfsrv_stablefirst.nsf_eograce)
4415 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4416 				NFSRV_LEASEDELTA;
4417 	}
4418 
4419 out:
4420 	NFSEXITCODE(error);
4421 	return (error);
4422 }
4423 
4424 /*
4425  * Do a server callback.
4426  * The "trunc" argument is slightly overloaded and refers to different
4427  * boolean arguments for CBRECALL and CBLAYOUTRECALL.
4428  */
4429 static int
4430 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp,
4431     int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp,
4432     int laytype, NFSPROC_T *p)
4433 {
4434 	struct mbuf *m;
4435 	u_int32_t *tl;
4436 	struct nfsrv_descript *nd;
4437 	struct ucred *cred;
4438 	int error = 0, slotpos;
4439 	u_int32_t callback;
4440 	struct nfsdsession *sep = NULL;
4441 	uint64_t tval;
4442 	bool dotls;
4443 
4444 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4445 	cred = newnfs_getcred();
4446 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4447 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4448 		NFSUNLOCKSTATE();
4449 		panic("docallb");
4450 	}
4451 	clp->lc_cbref++;
4452 
4453 	/*
4454 	 * Fill the callback program# and version into the request
4455 	 * structure for newnfs_connect() to use.
4456 	 */
4457 	clp->lc_req.nr_prog = clp->lc_program;
4458 #ifdef notnow
4459 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4460 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4461 	else
4462 #endif
4463 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4464 
4465 	/*
4466 	 * First, fill in some of the fields of nd and cr.
4467 	 */
4468 	nd->nd_flag = ND_NFSV4;
4469 	if (clp->lc_flags & LCL_GSS)
4470 		nd->nd_flag |= ND_KERBV;
4471 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4472 		nd->nd_flag |= ND_NFSV41;
4473 	if ((clp->lc_flags & LCL_NFSV42) != 0)
4474 		nd->nd_flag |= ND_NFSV42;
4475 	nd->nd_repstat = 0;
4476 	cred->cr_uid = clp->lc_uid;
4477 	cred->cr_gid = clp->lc_gid;
4478 	callback = clp->lc_callback;
4479 	NFSUNLOCKSTATE();
4480 	cred->cr_ngroups = 1;
4481 
4482 	/*
4483 	 * Get the first mbuf for the request.
4484 	 */
4485 	MGET(m, M_WAITOK, MT_DATA);
4486 	m->m_len = 0;
4487 	nd->nd_mreq = nd->nd_mb = m;
4488 	nd->nd_bpos = mtod(m, caddr_t);
4489 
4490 	/*
4491 	 * and build the callback request.
4492 	 */
4493 	if (procnum == NFSV4OP_CBGETATTR) {
4494 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4495 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4496 		    "CB Getattr", &sep, &slotpos);
4497 		if (error != 0) {
4498 			m_freem(nd->nd_mreq);
4499 			goto errout;
4500 		}
4501 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4502 		(void)nfsrv_putattrbit(nd, attrbitp);
4503 	} else if (procnum == NFSV4OP_CBRECALL) {
4504 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4505 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4506 		    "CB Recall", &sep, &slotpos);
4507 		if (error != 0) {
4508 			m_freem(nd->nd_mreq);
4509 			goto errout;
4510 		}
4511 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4512 		*tl++ = txdr_unsigned(stateidp->seqid);
4513 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4514 		    NFSX_STATEIDOTHER);
4515 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4516 		if (trunc)
4517 			*tl = newnfs_true;
4518 		else
4519 			*tl = newnfs_false;
4520 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4521 	} else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4522 		NFSD_DEBUG(4, "docallback layout recall\n");
4523 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4524 		error = nfsrv_cbcallargs(nd, clp, callback,
4525 		    NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep, &slotpos);
4526 		NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4527 		if (error != 0) {
4528 			m_freem(nd->nd_mreq);
4529 			goto errout;
4530 		}
4531 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4532 		*tl++ = txdr_unsigned(laytype);
4533 		*tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4534 		if (trunc)
4535 			*tl++ = newnfs_true;
4536 		else
4537 			*tl++ = newnfs_false;
4538 		*tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4539 		nfsm_fhtom(nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4540 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4541 		tval = 0;
4542 		txdr_hyper(tval, tl); tl += 2;
4543 		tval = UINT64_MAX;
4544 		txdr_hyper(tval, tl); tl += 2;
4545 		*tl++ = txdr_unsigned(stateidp->seqid);
4546 		NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4547 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4548 		NFSD_DEBUG(4, "aft args\n");
4549 	} else if (procnum == NFSV4PROC_CBNULL) {
4550 		nd->nd_procnum = NFSV4PROC_CBNULL;
4551 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4552 			error = nfsv4_getcbsession(clp, &sep);
4553 			if (error != 0) {
4554 				m_freem(nd->nd_mreq);
4555 				goto errout;
4556 			}
4557 		}
4558 	} else {
4559 		error = NFSERR_SERVERFAULT;
4560 		m_freem(nd->nd_mreq);
4561 		goto errout;
4562 	}
4563 
4564 	/*
4565 	 * Call newnfs_connect(), as required, and then newnfs_request().
4566 	 */
4567 	dotls = false;
4568 	if ((clp->lc_flags & LCL_TLSCB) != 0)
4569 		dotls = true;
4570 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4571 	if (clp->lc_req.nr_client == NULL) {
4572 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4573 			error = ECONNREFUSED;
4574 			if (procnum != NFSV4PROC_CBNULL)
4575 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4576 				    true);
4577 			nfsrv_freesession(sep, NULL);
4578 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4579 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4580 			    NULL, 1, dotls);
4581 		else
4582 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4583 			    NULL, 3, dotls);
4584 	}
4585 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4586 	NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4587 	if (!error) {
4588 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4589 			KASSERT(sep != NULL, ("sep NULL"));
4590 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4591 				error = newnfs_request(nd, NULL, clp,
4592 				    &clp->lc_req, NULL, NULL, cred,
4593 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4594 				    1, NULL, &sep->sess_cbsess);
4595 			else {
4596 				/*
4597 				 * This should probably never occur, but if a
4598 				 * client somehow does an RPC without a
4599 				 * SequenceID Op that causes a callback just
4600 				 * after the nfsd threads have been terminated
4601 				 * and restared we could conceivably get here
4602 				 * without a backchannel xprt.
4603 				 */
4604 				printf("nfsrv_docallback: no xprt\n");
4605 				error = ECONNREFUSED;
4606 			}
4607 			NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4608 			if (error != 0 && procnum != NFSV4PROC_CBNULL) {
4609 				/*
4610 				 * It is likely that the callback was never
4611 				 * processed by the client and, as such,
4612 				 * the sequence# for the session slot needs
4613 				 * to be backed up by one to avoid a
4614 				 * NFSERR_SEQMISORDERED error reply.
4615 				 * For the unlikely case where the callback
4616 				 * was processed by the client, this will
4617 				 * make the next callback on the slot
4618 				 * appear to be a retry.
4619 				 * Since callbacks never specify that the
4620 				 * reply be cached, this "apparent retry"
4621 				 * should not be a problem.
4622 				 */
4623 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4624 				    true);
4625 			}
4626 			nfsrv_freesession(sep, NULL);
4627 		} else
4628 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4629 			    NULL, NULL, cred, clp->lc_program,
4630 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4631 	}
4632 errout:
4633 	NFSFREECRED(cred);
4634 
4635 	/*
4636 	 * If error is set here, the Callback path isn't working
4637 	 * properly, so twiddle the appropriate LCL_ flags.
4638 	 * (nd_repstat != 0 indicates the Callback path is working,
4639 	 *  but the callback failed on the client.)
4640 	 */
4641 	if (error) {
4642 		/*
4643 		 * Mark the callback pathway down, which disabled issuing
4644 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4645 		 */
4646 		NFSLOCKSTATE();
4647 		clp->lc_flags |= LCL_CBDOWN;
4648 		NFSUNLOCKSTATE();
4649 	} else {
4650 		/*
4651 		 * Callback worked. If the callback path was down, disable
4652 		 * callbacks, so no more delegations will be issued. (This
4653 		 * is done on the assumption that the callback pathway is
4654 		 * flakey.)
4655 		 */
4656 		NFSLOCKSTATE();
4657 		if (clp->lc_flags & LCL_CBDOWN)
4658 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4659 		NFSUNLOCKSTATE();
4660 		if (nd->nd_repstat) {
4661 			error = nd->nd_repstat;
4662 			NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4663 			    procnum, error);
4664 		} else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4665 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4666 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4667 			    p, NULL);
4668 		m_freem(nd->nd_mrep);
4669 	}
4670 	NFSLOCKSTATE();
4671 	clp->lc_cbref--;
4672 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4673 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4674 		wakeup(clp);
4675 	}
4676 	NFSUNLOCKSTATE();
4677 
4678 	free(nd, M_TEMP);
4679 	NFSEXITCODE(error);
4680 	return (error);
4681 }
4682 
4683 /*
4684  * Set up the compound RPC for the callback.
4685  */
4686 static int
4687 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4688     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
4689     int *slotposp)
4690 {
4691 	uint32_t *tl;
4692 	int error, len;
4693 
4694 	len = strlen(optag);
4695 	(void)nfsm_strtom(nd, optag, len);
4696 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4697 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4698 		if ((nd->nd_flag & ND_NFSV42) != 0)
4699 			*tl++ = txdr_unsigned(NFSV42_MINORVERSION);
4700 		else
4701 			*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4702 		*tl++ = txdr_unsigned(callback);
4703 		*tl++ = txdr_unsigned(2);
4704 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4705 		error = nfsv4_setcbsequence(nd, clp, 1, sepp, slotposp);
4706 		if (error != 0)
4707 			return (error);
4708 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4709 		*tl = txdr_unsigned(op);
4710 	} else {
4711 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4712 		*tl++ = txdr_unsigned(callback);
4713 		*tl++ = txdr_unsigned(1);
4714 		*tl = txdr_unsigned(op);
4715 	}
4716 	return (0);
4717 }
4718 
4719 /*
4720  * Return the next index# for a clientid. Mostly just increment and return
4721  * the next one, but... if the 32bit unsigned does actually wrap around,
4722  * it should be rebooted.
4723  * At an average rate of one new client per second, it will wrap around in
4724  * approximately 136 years. (I think the server will have been shut
4725  * down or rebooted before then.)
4726  */
4727 static u_int32_t
4728 nfsrv_nextclientindex(void)
4729 {
4730 	static u_int32_t client_index = 0;
4731 
4732 	client_index++;
4733 	if (client_index != 0)
4734 		return (client_index);
4735 
4736 	printf("%s: out of clientids\n", __func__);
4737 	return (client_index);
4738 }
4739 
4740 /*
4741  * Return the next index# for a stateid. Mostly just increment and return
4742  * the next one, but... if the 32bit unsigned does actually wrap around
4743  * (will a BSD server stay up that long?), find
4744  * new start and end values.
4745  */
4746 static u_int32_t
4747 nfsrv_nextstateindex(struct nfsclient *clp)
4748 {
4749 	struct nfsstate *stp;
4750 	int i;
4751 	u_int32_t canuse, min_index, max_index;
4752 
4753 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4754 		clp->lc_stateindex++;
4755 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4756 			return (clp->lc_stateindex);
4757 	}
4758 
4759 	/*
4760 	 * Yuck, we've hit the end.
4761 	 * Look for a new min and max.
4762 	 */
4763 	min_index = 0;
4764 	max_index = 0xffffffff;
4765 	for (i = 0; i < nfsrv_statehashsize; i++) {
4766 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4767 		if (stp->ls_stateid.other[2] > 0x80000000) {
4768 		    if (stp->ls_stateid.other[2] < max_index)
4769 			max_index = stp->ls_stateid.other[2];
4770 		} else {
4771 		    if (stp->ls_stateid.other[2] > min_index)
4772 			min_index = stp->ls_stateid.other[2];
4773 		}
4774 	    }
4775 	}
4776 
4777 	/*
4778 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4779 	 */
4780 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4781 	    canuse = 0;
4782 	    /*
4783 	     * Loop around until we find an unused entry. Return that
4784 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4785 	     * (This is one of those rare cases where a goto is the
4786 	     *  cleanest way to code the loop.)
4787 	     */
4788 tryagain:
4789 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4790 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4791 		    if (stp->ls_stateid.other[2] == canuse) {
4792 			canuse++;
4793 			goto tryagain;
4794 		    }
4795 		}
4796 	    }
4797 	    clp->lc_flags |= LCL_INDEXNOTOK;
4798 	    return (canuse);
4799 	}
4800 
4801 	/*
4802 	 * Ok to start again from min + 1.
4803 	 */
4804 	clp->lc_stateindex = min_index + 1;
4805 	clp->lc_statemaxindex = max_index;
4806 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4807 	return (clp->lc_stateindex);
4808 }
4809 
4810 /*
4811  * The following functions handle the stable storage file that deals with
4812  * the edge conditions described in RFC3530 Sec. 8.6.3.
4813  * The file is as follows:
4814  * - a single record at the beginning that has the lease time of the
4815  *   previous server instance (before the last reboot) and the nfsrvboottime
4816  *   values for the previous server boots.
4817  *   These previous boot times are used to ensure that the current
4818  *   nfsrvboottime does not, somehow, get set to a previous one.
4819  *   (This is important so that Stale ClientIDs and StateIDs can
4820  *    be recognized.)
4821  *   The number of previous nfsvrboottime values precedes the list.
4822  * - followed by some number of appended records with:
4823  *   - client id string
4824  *   - flag that indicates it is a record revoking state via lease
4825  *     expiration or similar
4826  *     OR has successfully acquired state.
4827  * These structures vary in length, with the client string at the end, up
4828  * to NFSV4_OPAQUELIMIT in size.
4829  *
4830  * At the end of the grace period, the file is truncated, the first
4831  * record is rewritten with updated information and any acquired state
4832  * records for successful reclaims of state are written.
4833  *
4834  * Subsequent records are appended when the first state is issued to
4835  * a client and when state is revoked for a client.
4836  *
4837  * When reading the file in, state issued records that come later in
4838  * the file override older ones, since the append log is in cronological order.
4839  * If, for some reason, the file can't be read, the grace period is
4840  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4841  */
4842 
4843 /*
4844  * Read in the stable storage file. Called by nfssvc() before the nfsd
4845  * processes start servicing requests.
4846  */
4847 void
4848 nfsrv_setupstable(NFSPROC_T *p)
4849 {
4850 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4851 	struct nfsrv_stable *sp, *nsp;
4852 	struct nfst_rec *tsp;
4853 	int error, i, tryagain;
4854 	off_t off = 0;
4855 	ssize_t aresid, len;
4856 
4857 	/*
4858 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4859 	 * a reboot, so state has not been lost.
4860 	 */
4861 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4862 		return;
4863 	/*
4864 	 * Set Grace over just until the file reads successfully.
4865 	 */
4866 	nfsrvboottime = time_second;
4867 	LIST_INIT(&sf->nsf_head);
4868 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4869 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4870 	if (sf->nsf_fp == NULL)
4871 		return;
4872 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4873 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4874 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4875 	if (error || aresid || sf->nsf_numboots == 0 ||
4876 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4877 		return;
4878 
4879 	/*
4880 	 * Now, read in the boottimes.
4881 	 */
4882 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4883 		sizeof (time_t), M_TEMP, M_WAITOK);
4884 	off = sizeof (struct nfsf_rec);
4885 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4886 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4887 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4888 	if (error || aresid) {
4889 		free(sf->nsf_bootvals, M_TEMP);
4890 		sf->nsf_bootvals = NULL;
4891 		return;
4892 	}
4893 
4894 	/*
4895 	 * Make sure this nfsrvboottime is different from all recorded
4896 	 * previous ones.
4897 	 */
4898 	do {
4899 		tryagain = 0;
4900 		for (i = 0; i < sf->nsf_numboots; i++) {
4901 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4902 				nfsrvboottime++;
4903 				tryagain = 1;
4904 				break;
4905 			}
4906 		}
4907 	} while (tryagain);
4908 
4909 	sf->nsf_flags |= NFSNSF_OK;
4910 	off += (sf->nsf_numboots * sizeof (time_t));
4911 
4912 	/*
4913 	 * Read through the file, building a list of records for grace
4914 	 * checking.
4915 	 * Each record is between sizeof (struct nfst_rec) and
4916 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4917 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4918 	 */
4919 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4920 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4921 	do {
4922 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4923 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4924 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4925 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4926 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4927 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4928 		/*
4929 		 * Yuck, the file has been corrupted, so just return
4930 		 * after clearing out any restart state, so the grace period
4931 		 * is over.
4932 		 */
4933 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4934 			LIST_REMOVE(sp, nst_list);
4935 			free(sp, M_TEMP);
4936 		}
4937 		free(tsp, M_TEMP);
4938 		sf->nsf_flags &= ~NFSNSF_OK;
4939 		free(sf->nsf_bootvals, M_TEMP);
4940 		sf->nsf_bootvals = NULL;
4941 		return;
4942 	    }
4943 	    if (len > 0) {
4944 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4945 		/*
4946 		 * Search the list for a matching client.
4947 		 */
4948 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4949 			if (tsp->len == sp->nst_len &&
4950 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4951 				break;
4952 		}
4953 		if (sp == LIST_END(&sf->nsf_head)) {
4954 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4955 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4956 				M_WAITOK);
4957 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4958 				sizeof (struct nfst_rec) + tsp->len - 1);
4959 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4960 		} else {
4961 			if (tsp->flag == NFSNST_REVOKE)
4962 				sp->nst_flag |= NFSNST_REVOKE;
4963 			else
4964 				/*
4965 				 * A subsequent timestamp indicates the client
4966 				 * did a setclientid/confirm and any previous
4967 				 * revoke is no longer relevant.
4968 				 */
4969 				sp->nst_flag &= ~NFSNST_REVOKE;
4970 		}
4971 	    }
4972 	} while (len > 0);
4973 	free(tsp, M_TEMP);
4974 	sf->nsf_flags = NFSNSF_OK;
4975 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4976 		NFSRV_LEASEDELTA;
4977 }
4978 
4979 /*
4980  * Update the stable storage file, now that the grace period is over.
4981  */
4982 void
4983 nfsrv_updatestable(NFSPROC_T *p)
4984 {
4985 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4986 	struct nfsrv_stable *sp, *nsp;
4987 	int i;
4988 	struct nfsvattr nva;
4989 	vnode_t vp;
4990 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4991 	mount_t mp = NULL;
4992 #endif
4993 	int error;
4994 
4995 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4996 		return;
4997 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4998 	/*
4999 	 * Ok, we need to rewrite the stable storage file.
5000 	 * - truncate to 0 length
5001 	 * - write the new first structure
5002 	 * - loop through the data structures, writing out any that
5003 	 *   have timestamps older than the old boot
5004 	 */
5005 	if (sf->nsf_bootvals) {
5006 		sf->nsf_numboots++;
5007 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
5008 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
5009 	} else {
5010 		sf->nsf_numboots = 1;
5011 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
5012 			M_TEMP, M_WAITOK);
5013 	}
5014 	sf->nsf_bootvals[0] = nfsrvboottime;
5015 	sf->nsf_lease = nfsrv_lease;
5016 	NFSVNO_ATTRINIT(&nva);
5017 	NFSVNO_SETATTRVAL(&nva, size, 0);
5018 	vp = NFSFPVNODE(sf->nsf_fp);
5019 	vn_start_write(vp, &mp, V_WAIT);
5020 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5021 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
5022 		    NULL);
5023 		NFSVOPUNLOCK(vp);
5024 	} else
5025 		error = EPERM;
5026 	vn_finished_write(mp);
5027 	if (!error)
5028 	    error = NFSD_RDWR(UIO_WRITE, vp,
5029 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
5030 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5031 	if (!error)
5032 	    error = NFSD_RDWR(UIO_WRITE, vp,
5033 		(caddr_t)sf->nsf_bootvals,
5034 		sf->nsf_numboots * sizeof (time_t),
5035 		(off_t)(sizeof (struct nfsf_rec)),
5036 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5037 	free(sf->nsf_bootvals, M_TEMP);
5038 	sf->nsf_bootvals = NULL;
5039 	if (error) {
5040 		sf->nsf_flags &= ~NFSNSF_OK;
5041 		printf("EEK! Can't write NfsV4 stable storage file\n");
5042 		return;
5043 	}
5044 	sf->nsf_flags |= NFSNSF_OK;
5045 
5046 	/*
5047 	 * Loop through the list and write out timestamp records for
5048 	 * any clients that successfully reclaimed state.
5049 	 */
5050 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
5051 		if (sp->nst_flag & NFSNST_GOTSTATE) {
5052 			nfsrv_writestable(sp->nst_client, sp->nst_len,
5053 				NFSNST_NEWSTATE, p);
5054 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
5055 		}
5056 		LIST_REMOVE(sp, nst_list);
5057 		free(sp, M_TEMP);
5058 	}
5059 	nfsrv_backupstable();
5060 }
5061 
5062 /*
5063  * Append a record to the stable storage file.
5064  */
5065 void
5066 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
5067 {
5068 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
5069 	struct nfst_rec *sp;
5070 	int error;
5071 
5072 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
5073 		return;
5074 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
5075 		len - 1, M_TEMP, M_WAITOK);
5076 	sp->len = len;
5077 	NFSBCOPY(client, sp->client, len);
5078 	sp->flag = flag;
5079 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
5080 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
5081 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
5082 	free(sp, M_TEMP);
5083 	if (error) {
5084 		sf->nsf_flags &= ~NFSNSF_OK;
5085 		printf("EEK! Can't write NfsV4 stable storage file\n");
5086 	}
5087 }
5088 
5089 /*
5090  * This function is called during the grace period to mark a client
5091  * that successfully reclaimed state.
5092  */
5093 static void
5094 nfsrv_markstable(struct nfsclient *clp)
5095 {
5096 	struct nfsrv_stable *sp;
5097 
5098 	/*
5099 	 * First find the client structure.
5100 	 */
5101 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5102 		if (sp->nst_len == clp->lc_idlen &&
5103 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5104 			break;
5105 	}
5106 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
5107 		return;
5108 
5109 	/*
5110 	 * Now, just mark it and set the nfsclient back pointer.
5111 	 */
5112 	sp->nst_flag |= NFSNST_GOTSTATE;
5113 	sp->nst_clp = clp;
5114 }
5115 
5116 /*
5117  * This function is called when a NFSv4.1 client does a ReclaimComplete.
5118  * Very similar to nfsrv_markstable(), except for the flag being set.
5119  */
5120 static void
5121 nfsrv_markreclaim(struct nfsclient *clp)
5122 {
5123 	struct nfsrv_stable *sp;
5124 
5125 	/*
5126 	 * First find the client structure.
5127 	 */
5128 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5129 		if (sp->nst_len == clp->lc_idlen &&
5130 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5131 			break;
5132 	}
5133 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
5134 		return;
5135 
5136 	/*
5137 	 * Now, just set the flag.
5138 	 */
5139 	sp->nst_flag |= NFSNST_RECLAIMED;
5140 }
5141 
5142 /*
5143  * This function is called for a reclaim, to see if it gets grace.
5144  * It returns 0 if a reclaim is allowed, 1 otherwise.
5145  */
5146 static int
5147 nfsrv_checkstable(struct nfsclient *clp)
5148 {
5149 	struct nfsrv_stable *sp;
5150 
5151 	/*
5152 	 * First, find the entry for the client.
5153 	 */
5154 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5155 		if (sp->nst_len == clp->lc_idlen &&
5156 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5157 			break;
5158 	}
5159 
5160 	/*
5161 	 * If not in the list, state was revoked or no state was issued
5162 	 * since the previous reboot, a reclaim is denied.
5163 	 */
5164 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
5165 	    (sp->nst_flag & NFSNST_REVOKE) ||
5166 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
5167 		return (1);
5168 	return (0);
5169 }
5170 
5171 /*
5172  * Test for and try to clear out a conflicting client. This is called by
5173  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5174  * a found.
5175  * The trick here is that it can't revoke a conflicting client with an
5176  * expired lease unless it holds the v4root lock, so...
5177  * If no v4root lock, get the lock and return 1 to indicate "try again".
5178  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5179  * the revocation worked and the conflicting client is "bye, bye", so it
5180  * can be tried again.
5181  * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK().
5182  * Unlocks State before a non-zero value is returned.
5183  */
5184 static int
5185 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5186     NFSPROC_T *p)
5187 {
5188 	int gotlock, lktype = 0;
5189 
5190 	/*
5191 	 * If lease hasn't expired, we can't fix it.
5192 	 */
5193 	if (clp->lc_expiry >= NFSD_MONOSEC ||
5194 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
5195 		return (0);
5196 	if (*haslockp == 0) {
5197 		NFSUNLOCKSTATE();
5198 		if (vp != NULL) {
5199 			lktype = NFSVOPISLOCKED(vp);
5200 			NFSVOPUNLOCK(vp);
5201 		}
5202 		NFSLOCKV4ROOTMUTEX();
5203 		nfsv4_relref(&nfsv4rootfs_lock);
5204 		do {
5205 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5206 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5207 		} while (!gotlock);
5208 		NFSUNLOCKV4ROOTMUTEX();
5209 		*haslockp = 1;
5210 		if (vp != NULL) {
5211 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5212 			if (VN_IS_DOOMED(vp))
5213 				return (2);
5214 		}
5215 		return (1);
5216 	}
5217 	NFSUNLOCKSTATE();
5218 
5219 	/*
5220 	 * Ok, we can expire the conflicting client.
5221 	 */
5222 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5223 	nfsrv_backupstable();
5224 	nfsrv_cleanclient(clp, p);
5225 	nfsrv_freedeleglist(&clp->lc_deleg);
5226 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5227 	LIST_REMOVE(clp, lc_hash);
5228 	nfsrv_zapclient(clp, p);
5229 	return (1);
5230 }
5231 
5232 /*
5233  * Resolve a delegation conflict.
5234  * Returns 0 to indicate the conflict was resolved without sleeping.
5235  * Return -1 to indicate that the caller should check for conflicts again.
5236  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5237  *
5238  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5239  * for a return of 0, since there was no sleep and it could be required
5240  * later. It is released for a return of NFSERR_DELAY, since the caller
5241  * will return that error. It is released when a sleep was done waiting
5242  * for the delegation to be returned or expire (so that other nfsds can
5243  * handle ops). Then, it must be acquired for the write to stable storage.
5244  * (This function is somewhat similar to nfsrv_clientconflict(), but
5245  *  the semantics differ in a couple of subtle ways. The return of 0
5246  *  indicates the conflict was resolved without sleeping here, not
5247  *  that the conflict can't be resolved and the handling of nfsv4root_lock
5248  *  differs, as noted above.)
5249  * Unlocks State before returning a non-zero value.
5250  */
5251 static int
5252 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5253     vnode_t vp)
5254 {
5255 	struct nfsclient *clp = stp->ls_clp;
5256 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5257 	nfsv4stateid_t tstateid;
5258 	fhandle_t tfh;
5259 
5260 	/*
5261 	 * If the conflict is with an old delegation...
5262 	 */
5263 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
5264 		/*
5265 		 * You can delete it, if it has expired.
5266 		 */
5267 		if (clp->lc_delegtime < NFSD_MONOSEC) {
5268 			nfsrv_freedeleg(stp);
5269 			NFSUNLOCKSTATE();
5270 			error = -1;
5271 			goto out;
5272 		}
5273 		NFSUNLOCKSTATE();
5274 		/*
5275 		 * During this delay, the old delegation could expire or it
5276 		 * could be recovered by the client via an Open with
5277 		 * CLAIM_DELEGATE_PREV.
5278 		 * Release the nfsv4root_lock, if held.
5279 		 */
5280 		if (*haslockp) {
5281 			*haslockp = 0;
5282 			NFSLOCKV4ROOTMUTEX();
5283 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5284 			NFSUNLOCKV4ROOTMUTEX();
5285 		}
5286 		error = NFSERR_DELAY;
5287 		goto out;
5288 	}
5289 
5290 	/*
5291 	 * It's a current delegation, so:
5292 	 * - check to see if the delegation has expired
5293 	 *   - if so, get the v4root lock and then expire it
5294 	 */
5295 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0 || (stp->ls_lastrecall <
5296 	    NFSD_MONOSEC && clp->lc_expiry >= NFSD_MONOSEC &&
5297 	    stp->ls_delegtime >= NFSD_MONOSEC)) {
5298 		/*
5299 		 * - do a recall callback, since not yet done
5300 		 * For now, never allow truncate to be set. To use
5301 		 * truncate safely, it must be guaranteed that the
5302 		 * Remove, Rename or Setattr with size of 0 will
5303 		 * succeed and that would require major changes to
5304 		 * the VFS/Vnode OPs.
5305 		 * Set the expiry time large enough so that it won't expire
5306 		 * until after the callback, then set it correctly, once
5307 		 * the callback is done. (The delegation will now time
5308 		 * out whether or not the Recall worked ok. The timeout
5309 		 * will be extended when ops are done on the delegation
5310 		 * stateid, up to the timelimit.)
5311 		 */
5312 		if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) {
5313 			stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5314 			    NFSRV_LEASEDELTA;
5315 			stp->ls_delegtimelimit = NFSD_MONOSEC + (6 *
5316 			    nfsrv_lease) + NFSRV_LEASEDELTA;
5317 			stp->ls_flags |= NFSLCK_DELEGRECALL;
5318 		}
5319 		stp->ls_lastrecall = time_uptime + 1;
5320 
5321 		/*
5322 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5323 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5324 		 * in order to try and avoid a race that could happen
5325 		 * when a CBRecall request passed the Open reply with
5326 		 * the delegation in it when transitting the network.
5327 		 * Since nfsrv_docallback will sleep, don't use stp after
5328 		 * the call.
5329 		 */
5330 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5331 		    sizeof (tstateid));
5332 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5333 		    sizeof (tfh));
5334 		NFSUNLOCKSTATE();
5335 		if (*haslockp) {
5336 			*haslockp = 0;
5337 			NFSLOCKV4ROOTMUTEX();
5338 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5339 			NFSUNLOCKV4ROOTMUTEX();
5340 		}
5341 		retrycnt = 0;
5342 		do {
5343 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5344 			&tstateid, 0, &tfh, NULL, NULL, 0, p);
5345 		    retrycnt++;
5346 		} while ((error == NFSERR_BADSTATEID ||
5347 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5348 		error = NFSERR_DELAY;
5349 		goto out;
5350 	}
5351 
5352 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5353 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5354 		NFSUNLOCKSTATE();
5355 		/*
5356 		 * A recall has been done, but it has not yet expired.
5357 		 * So, RETURN_DELAY.
5358 		 */
5359 		if (*haslockp) {
5360 			*haslockp = 0;
5361 			NFSLOCKV4ROOTMUTEX();
5362 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5363 			NFSUNLOCKV4ROOTMUTEX();
5364 		}
5365 		error = NFSERR_DELAY;
5366 		goto out;
5367 	}
5368 
5369 	/*
5370 	 * If we don't yet have the lock, just get it and then return,
5371 	 * since we need that before deleting expired state, such as
5372 	 * this delegation.
5373 	 * When getting the lock, unlock the vnode, so other nfsds that
5374 	 * are in progress, won't get stuck waiting for the vnode lock.
5375 	 */
5376 	if (*haslockp == 0) {
5377 		NFSUNLOCKSTATE();
5378 		if (vp != NULL) {
5379 			lktype = NFSVOPISLOCKED(vp);
5380 			NFSVOPUNLOCK(vp);
5381 		}
5382 		NFSLOCKV4ROOTMUTEX();
5383 		nfsv4_relref(&nfsv4rootfs_lock);
5384 		do {
5385 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5386 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5387 		} while (!gotlock);
5388 		NFSUNLOCKV4ROOTMUTEX();
5389 		*haslockp = 1;
5390 		if (vp != NULL) {
5391 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5392 			if (VN_IS_DOOMED(vp)) {
5393 				*haslockp = 0;
5394 				NFSLOCKV4ROOTMUTEX();
5395 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5396 				NFSUNLOCKV4ROOTMUTEX();
5397 				error = NFSERR_PERM;
5398 				goto out;
5399 			}
5400 		}
5401 		error = -1;
5402 		goto out;
5403 	}
5404 
5405 	NFSUNLOCKSTATE();
5406 	/*
5407 	 * Ok, we can delete the expired delegation.
5408 	 * First, write the Revoke record to stable storage and then
5409 	 * clear out the conflict.
5410 	 * Since all other nfsd threads are now blocked, we can safely
5411 	 * sleep without the state changing.
5412 	 */
5413 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5414 	nfsrv_backupstable();
5415 	if (clp->lc_expiry < NFSD_MONOSEC) {
5416 		nfsrv_cleanclient(clp, p);
5417 		nfsrv_freedeleglist(&clp->lc_deleg);
5418 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5419 		LIST_REMOVE(clp, lc_hash);
5420 		zapped_clp = 1;
5421 	} else {
5422 		nfsrv_freedeleg(stp);
5423 		zapped_clp = 0;
5424 	}
5425 	if (zapped_clp)
5426 		nfsrv_zapclient(clp, p);
5427 	error = -1;
5428 
5429 out:
5430 	NFSEXITCODE(error);
5431 	return (error);
5432 }
5433 
5434 /*
5435  * Check for a remove allowed, if remove is set to 1 and get rid of
5436  * delegations.
5437  */
5438 int
5439 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd,
5440     nfsquad_t clientid, NFSPROC_T *p)
5441 {
5442 	struct nfsclient *clp;
5443 	struct nfsstate *stp;
5444 	struct nfslockfile *lfp;
5445 	int error, haslock = 0;
5446 	fhandle_t nfh;
5447 
5448 	clp = NULL;
5449 	/*
5450 	 * First, get the lock file structure.
5451 	 * (A return of -1 means no associated state, so remove ok.)
5452 	 */
5453 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5454 tryagain:
5455 	NFSLOCKSTATE();
5456 	if (error == 0 && clientid.qval != 0)
5457 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
5458 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
5459 	if (!error)
5460 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5461 	if (error) {
5462 		NFSUNLOCKSTATE();
5463 		if (haslock) {
5464 			NFSLOCKV4ROOTMUTEX();
5465 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5466 			NFSUNLOCKV4ROOTMUTEX();
5467 		}
5468 		if (error == -1)
5469 			error = 0;
5470 		goto out;
5471 	}
5472 
5473 	/*
5474 	 * Now, we must Recall any delegations.
5475 	 */
5476 	error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
5477 	if (error) {
5478 		/*
5479 		 * nfsrv_cleandeleg() unlocks state for non-zero
5480 		 * return.
5481 		 */
5482 		if (error == -1)
5483 			goto tryagain;
5484 		if (haslock) {
5485 			NFSLOCKV4ROOTMUTEX();
5486 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5487 			NFSUNLOCKV4ROOTMUTEX();
5488 		}
5489 		goto out;
5490 	}
5491 
5492 	/*
5493 	 * Now, look for a conflicting open share.
5494 	 */
5495 	if (remove) {
5496 		/*
5497 		 * If the entry in the directory was the last reference to the
5498 		 * corresponding filesystem object, the object can be destroyed
5499 		 * */
5500 		if(lfp->lf_usecount>1)
5501 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5502 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5503 					error = NFSERR_FILEOPEN;
5504 					break;
5505 				}
5506 			}
5507 	}
5508 
5509 	NFSUNLOCKSTATE();
5510 	if (haslock) {
5511 		NFSLOCKV4ROOTMUTEX();
5512 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5513 		NFSUNLOCKV4ROOTMUTEX();
5514 	}
5515 
5516 out:
5517 	NFSEXITCODE(error);
5518 	return (error);
5519 }
5520 
5521 /*
5522  * Clear out all delegations for the file referred to by lfp.
5523  * May return NFSERR_DELAY, if there will be a delay waiting for
5524  * delegations to expire.
5525  * Returns -1 to indicate it slept while recalling a delegation.
5526  * This function has the side effect of deleting the nfslockfile structure,
5527  * if it no longer has associated state and didn't have to sleep.
5528  * Unlocks State before a non-zero value is returned.
5529  */
5530 static int
5531 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5532     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5533 {
5534 	struct nfsstate *stp, *nstp;
5535 	int ret = 0;
5536 
5537 	stp = LIST_FIRST(&lfp->lf_deleg);
5538 	while (stp != LIST_END(&lfp->lf_deleg)) {
5539 		nstp = LIST_NEXT(stp, ls_file);
5540 		if (stp->ls_clp != clp) {
5541 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5542 			if (ret) {
5543 				/*
5544 				 * nfsrv_delegconflict() unlocks state
5545 				 * when it returns non-zero.
5546 				 */
5547 				goto out;
5548 			}
5549 		}
5550 		stp = nstp;
5551 	}
5552 out:
5553 	NFSEXITCODE(ret);
5554 	return (ret);
5555 }
5556 
5557 /*
5558  * There are certain operations that, when being done outside of NFSv4,
5559  * require that any NFSv4 delegation for the file be recalled.
5560  * This function is to be called for those cases:
5561  * VOP_RENAME() - When a delegation is being recalled for any reason,
5562  *	the client may have to do Opens against the server, using the file's
5563  *	final component name. If the file has been renamed on the server,
5564  *	that component name will be incorrect and the Open will fail.
5565  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5566  *	been removed on the server, if there is a delegation issued to
5567  *	that client for the file. I say "theoretically" since clients
5568  *	normally do an Access Op before the Open and that Access Op will
5569  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5570  *	they will detect the file's removal in the same manner. (There is
5571  *	one case where RFC3530 allows a client to do an Open without first
5572  *	doing an Access Op, which is passage of a check against the ACE
5573  *	returned with a Write delegation, but current practice is to ignore
5574  *	the ACE and always do an Access Op.)
5575  *	Since the functions can only be called with an unlocked vnode, this
5576  *	can't be done at this time.
5577  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5578  *	locks locally in the client, which are not visible to the server. To
5579  *	deal with this, issuing of delegations for a vnode must be disabled
5580  *	and all delegations for the vnode recalled. This is done via the
5581  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5582  */
5583 void
5584 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5585 {
5586 	time_t starttime;
5587 	int error;
5588 
5589 	/*
5590 	 * First, check to see if the server is currently running and it has
5591 	 * been called for a regular file when issuing delegations.
5592 	 */
5593 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5594 	    nfsrv_issuedelegs == 0)
5595 		return;
5596 
5597 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5598 	/*
5599 	 * First, get a reference on the nfsv4rootfs_lock so that an
5600 	 * exclusive lock cannot be acquired by another thread.
5601 	 */
5602 	NFSLOCKV4ROOTMUTEX();
5603 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5604 	NFSUNLOCKV4ROOTMUTEX();
5605 
5606 	/*
5607 	 * Now, call nfsrv_checkremove() in a loop while it returns
5608 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5609 	 */
5610 	starttime = NFSD_MONOSEC;
5611 	do {
5612 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5613 			error = nfsrv_checkremove(vp, 0, NULL,
5614 			    (nfsquad_t)((u_quad_t)0), p);
5615 			NFSVOPUNLOCK(vp);
5616 		} else
5617 			error = EPERM;
5618 		if (error == NFSERR_DELAY) {
5619 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5620 				break;
5621 			/* Sleep for a short period of time */
5622 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5623 		}
5624 	} while (error == NFSERR_DELAY);
5625 	NFSLOCKV4ROOTMUTEX();
5626 	nfsv4_relref(&nfsv4rootfs_lock);
5627 	NFSUNLOCKV4ROOTMUTEX();
5628 }
5629 
5630 void
5631 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5632 {
5633 
5634 #ifdef VV_DISABLEDELEG
5635 	/*
5636 	 * First, flag issuance of delegations disabled.
5637 	 */
5638 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5639 #endif
5640 
5641 	/*
5642 	 * Then call nfsd_recalldelegation() to get rid of all extant
5643 	 * delegations.
5644 	 */
5645 	nfsd_recalldelegation(vp, p);
5646 }
5647 
5648 /*
5649  * Check for conflicting locks, etc. and then get rid of delegations.
5650  * (At one point I thought that I should get rid of delegations for any
5651  *  Setattr, since it could potentially disallow the I/O op (read or write)
5652  *  allowed by the delegation. However, Setattr Ops that aren't changing
5653  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5654  *  for the same client or a different one, so I decided to only get rid
5655  *  of delegations for other clients when the size is being changed.)
5656  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5657  * as Write backs, even if there is no delegation, so it really isn't any
5658  * different?)
5659  */
5660 int
5661 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5662     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5663     struct nfsexstuff *exp, NFSPROC_T *p)
5664 {
5665 	struct nfsstate st, *stp = &st;
5666 	struct nfslock lo, *lop = &lo;
5667 	int error = 0;
5668 	nfsquad_t clientid;
5669 
5670 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5671 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5672 		lop->lo_first = nvap->na_size;
5673 	} else {
5674 		stp->ls_flags = 0;
5675 		lop->lo_first = 0;
5676 	}
5677 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5678 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5679 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5680 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5681 		stp->ls_flags |= NFSLCK_SETATTR;
5682 	if (stp->ls_flags == 0)
5683 		goto out;
5684 	lop->lo_end = NFS64BITSSET;
5685 	lop->lo_flags = NFSLCK_WRITE;
5686 	stp->ls_ownerlen = 0;
5687 	stp->ls_op = NULL;
5688 	stp->ls_uid = nd->nd_cred->cr_uid;
5689 	stp->ls_stateid.seqid = stateidp->seqid;
5690 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5691 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5692 	stp->ls_stateid.other[2] = stateidp->other[2];
5693 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5694 	    stateidp, exp, nd, p);
5695 
5696 out:
5697 	NFSEXITCODE2(error, nd);
5698 	return (error);
5699 }
5700 
5701 /*
5702  * Check for a write delegation and do a CBGETATTR if there is one, updating
5703  * the attributes, as required.
5704  * Should I return an error if I can't get the attributes? (For now, I'll
5705  * just return ok.
5706  */
5707 int
5708 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5709     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5710 {
5711 	struct nfsstate *stp;
5712 	struct nfslockfile *lfp;
5713 	struct nfsclient *clp;
5714 	struct nfsvattr nva;
5715 	fhandle_t nfh;
5716 	int error = 0;
5717 	nfsattrbit_t cbbits;
5718 	u_quad_t delegfilerev;
5719 
5720 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5721 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5722 		goto out;
5723 	if (nfsrv_writedelegcnt == 0)
5724 		goto out;
5725 
5726 	/*
5727 	 * Get the lock file structure.
5728 	 * (A return of -1 means no associated state, so return ok.)
5729 	 */
5730 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5731 	NFSLOCKSTATE();
5732 	if (!error)
5733 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5734 	if (error) {
5735 		NFSUNLOCKSTATE();
5736 		if (error == -1)
5737 			error = 0;
5738 		goto out;
5739 	}
5740 
5741 	/*
5742 	 * Now, look for a write delegation.
5743 	 */
5744 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5745 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5746 			break;
5747 	}
5748 	if (stp == LIST_END(&lfp->lf_deleg)) {
5749 		NFSUNLOCKSTATE();
5750 		goto out;
5751 	}
5752 	clp = stp->ls_clp;
5753 
5754 	/* If the clientid is not confirmed, ignore the delegation. */
5755 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
5756 		NFSUNLOCKSTATE();
5757 		goto out;
5758 	}
5759 
5760 	delegfilerev = stp->ls_filerev;
5761 	/*
5762 	 * If the Write delegation was issued as a part of this Compound RPC
5763 	 * or if we have an Implied Clientid (used in a previous Op in this
5764 	 * compound) and it is the client the delegation was issued to,
5765 	 * just return ok.
5766 	 * I also assume that it is from the same client iff the network
5767 	 * host IP address is the same as the callback address. (Not
5768 	 * exactly correct by the RFC, but avoids a lot of Getattr
5769 	 * callbacks.)
5770 	 */
5771 	if (nd->nd_compref == stp->ls_compref ||
5772 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5773 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5774 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5775 		NFSUNLOCKSTATE();
5776 		goto out;
5777 	}
5778 
5779 	/*
5780 	 * We are now done with the delegation state structure,
5781 	 * so the statelock can be released and we can now tsleep().
5782 	 */
5783 
5784 	/*
5785 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5786 	 * has changed.
5787 	 */
5788 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5789 		NFSUNLOCKSTATE();
5790 		NFSVNO_ATTRINIT(&nva);
5791 		nva.na_filerev = NFS64BITSSET;
5792 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5793 		    0, &nfh, &nva, &cbbits, 0, p);
5794 		if (!error) {
5795 			if ((nva.na_filerev != NFS64BITSSET &&
5796 			    nva.na_filerev > delegfilerev) ||
5797 			    (NFSVNO_ISSETSIZE(&nva) &&
5798 			     nva.na_size != nvap->na_size)) {
5799 				error = nfsvno_updfilerev(vp, nvap, nd, p);
5800 				if (NFSVNO_ISSETSIZE(&nva))
5801 					nvap->na_size = nva.na_size;
5802 			}
5803 		} else
5804 			error = 0;	/* Ignore callback errors for now. */
5805 	} else {
5806 		NFSUNLOCKSTATE();
5807 	}
5808 
5809 out:
5810 	NFSEXITCODE2(error, nd);
5811 	return (error);
5812 }
5813 
5814 /*
5815  * This function looks for openowners that haven't had any opens for
5816  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5817  * is set.
5818  */
5819 void
5820 nfsrv_throwawayopens(NFSPROC_T *p)
5821 {
5822 	struct nfsclient *clp, *nclp;
5823 	struct nfsstate *stp, *nstp;
5824 	int i;
5825 
5826 	NFSLOCKSTATE();
5827 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5828 	/*
5829 	 * For each client...
5830 	 */
5831 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5832 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5833 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5834 			if (LIST_EMPTY(&stp->ls_open) &&
5835 			    (stp->ls_noopens > NFSNOOPEN ||
5836 			     (nfsrv_openpluslock * 2) >
5837 			     nfsrv_v4statelimit))
5838 				nfsrv_freeopenowner(stp, 0, p);
5839 		}
5840 	    }
5841 	}
5842 	NFSUNLOCKSTATE();
5843 }
5844 
5845 /*
5846  * This function checks to see if the credentials are the same.
5847  * Returns 1 for not same, 0 otherwise.
5848  */
5849 static int
5850 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5851 {
5852 
5853 	if (nd->nd_flag & ND_GSS) {
5854 		if (!(clp->lc_flags & LCL_GSS))
5855 			return (1);
5856 		if (clp->lc_flags & LCL_NAME) {
5857 			if (nd->nd_princlen != clp->lc_namelen ||
5858 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5859 				clp->lc_namelen))
5860 				return (1);
5861 			else
5862 				return (0);
5863 		}
5864 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5865 			return (0);
5866 		else
5867 			return (1);
5868 	} else if (clp->lc_flags & LCL_GSS)
5869 		return (1);
5870 	/*
5871 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5872 	 * in RFC3530, which talks about principals, but doesn't say anything
5873 	 * about uids for AUTH_SYS.)
5874 	 */
5875 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5876 		return (0);
5877 	else
5878 		return (1);
5879 }
5880 
5881 /*
5882  * Calculate the lease expiry time.
5883  */
5884 static time_t
5885 nfsrv_leaseexpiry(void)
5886 {
5887 
5888 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5889 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5890 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5891 }
5892 
5893 /*
5894  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5895  */
5896 static void
5897 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5898 {
5899 
5900 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5901 		return;
5902 
5903 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5904 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5905 		stp->ls_delegtime += nfsrv_lease;
5906 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5907 			stp->ls_delegtime = stp->ls_delegtimelimit;
5908 	}
5909 }
5910 
5911 /*
5912  * This function checks to see if there is any other state associated
5913  * with the openowner for this Open.
5914  * It returns 1 if there is no other state, 0 otherwise.
5915  */
5916 static int
5917 nfsrv_nootherstate(struct nfsstate *stp)
5918 {
5919 	struct nfsstate *tstp;
5920 
5921 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5922 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5923 			return (0);
5924 	}
5925 	return (1);
5926 }
5927 
5928 /*
5929  * Create a list of lock deltas (changes to local byte range locking
5930  * that can be rolled back using the list) and apply the changes via
5931  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5932  * the rollback or update function will be called after this.
5933  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5934  * call fails. If it returns an error, it will unlock the list.
5935  */
5936 static int
5937 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5938     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5939 {
5940 	struct nfslock *lop, *nlop;
5941 	int error = 0;
5942 
5943 	/* Loop through the list of locks. */
5944 	lop = LIST_FIRST(&lfp->lf_locallock);
5945 	while (first < end && lop != NULL) {
5946 		nlop = LIST_NEXT(lop, lo_lckowner);
5947 		if (first >= lop->lo_end) {
5948 			/* not there yet */
5949 			lop = nlop;
5950 		} else if (first < lop->lo_first) {
5951 			/* new one starts before entry in list */
5952 			if (end <= lop->lo_first) {
5953 				/* no overlap between old and new */
5954 				error = nfsrv_dolocal(vp, lfp, flags,
5955 				    NFSLCK_UNLOCK, first, end, cfp, p);
5956 				if (error != 0)
5957 					break;
5958 				first = end;
5959 			} else {
5960 				/* handle fragment overlapped with new one */
5961 				error = nfsrv_dolocal(vp, lfp, flags,
5962 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5963 				    p);
5964 				if (error != 0)
5965 					break;
5966 				first = lop->lo_first;
5967 			}
5968 		} else {
5969 			/* new one overlaps this entry in list */
5970 			if (end <= lop->lo_end) {
5971 				/* overlaps all of new one */
5972 				error = nfsrv_dolocal(vp, lfp, flags,
5973 				    lop->lo_flags, first, end, cfp, p);
5974 				if (error != 0)
5975 					break;
5976 				first = end;
5977 			} else {
5978 				/* handle fragment overlapped with new one */
5979 				error = nfsrv_dolocal(vp, lfp, flags,
5980 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5981 				if (error != 0)
5982 					break;
5983 				first = lop->lo_end;
5984 				lop = nlop;
5985 			}
5986 		}
5987 	}
5988 	if (first < end && error == 0)
5989 		/* handle fragment past end of list */
5990 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5991 		    end, cfp, p);
5992 
5993 	NFSEXITCODE(error);
5994 	return (error);
5995 }
5996 
5997 /*
5998  * Local lock unlock. Unlock all byte ranges that are no longer locked
5999  * by NFSv4. To do this, unlock any subranges of first-->end that
6000  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
6001  * list. This list has all locks for the file held by other
6002  * <clientid, lockowner> tuples. The list is ordered by increasing
6003  * lo_first value, but may have entries that overlap each other, for
6004  * the case of read locks.
6005  */
6006 static void
6007 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
6008     uint64_t init_end, NFSPROC_T *p)
6009 {
6010 	struct nfslock *lop;
6011 	uint64_t first, end, prevfirst __unused;
6012 
6013 	first = init_first;
6014 	end = init_end;
6015 	while (first < init_end) {
6016 		/* Loop through all nfs locks, adjusting first and end */
6017 		prevfirst = 0;
6018 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
6019 			KASSERT(prevfirst <= lop->lo_first,
6020 			    ("nfsv4 locks out of order"));
6021 			KASSERT(lop->lo_first < lop->lo_end,
6022 			    ("nfsv4 bogus lock"));
6023 			prevfirst = lop->lo_first;
6024 			if (first >= lop->lo_first &&
6025 			    first < lop->lo_end)
6026 				/*
6027 				 * Overlaps with initial part, so trim
6028 				 * off that initial part by moving first past
6029 				 * it.
6030 				 */
6031 				first = lop->lo_end;
6032 			else if (end > lop->lo_first &&
6033 			    lop->lo_first > first) {
6034 				/*
6035 				 * This lock defines the end of the
6036 				 * segment to unlock, so set end to the
6037 				 * start of it and break out of the loop.
6038 				 */
6039 				end = lop->lo_first;
6040 				break;
6041 			}
6042 			if (first >= end)
6043 				/*
6044 				 * There is no segment left to do, so
6045 				 * break out of this loop and then exit
6046 				 * the outer while() since first will be set
6047 				 * to end, which must equal init_end here.
6048 				 */
6049 				break;
6050 		}
6051 		if (first < end) {
6052 			/* Unlock this segment */
6053 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
6054 			    NFSLCK_READ, first, end, NULL, p);
6055 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
6056 			    first, end);
6057 		}
6058 		/*
6059 		 * Now move past this segment and look for any further
6060 		 * segment in the range, if there is one.
6061 		 */
6062 		first = end;
6063 		end = init_end;
6064 	}
6065 }
6066 
6067 /*
6068  * Do the local lock operation and update the rollback list, as required.
6069  * Perform the rollback and return the error if nfsvno_advlock() fails.
6070  */
6071 static int
6072 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
6073     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
6074 {
6075 	struct nfsrollback *rlp;
6076 	int error = 0, ltype, oldltype;
6077 
6078 	if (flags & NFSLCK_WRITE)
6079 		ltype = F_WRLCK;
6080 	else if (flags & NFSLCK_READ)
6081 		ltype = F_RDLCK;
6082 	else
6083 		ltype = F_UNLCK;
6084 	if (oldflags & NFSLCK_WRITE)
6085 		oldltype = F_WRLCK;
6086 	else if (oldflags & NFSLCK_READ)
6087 		oldltype = F_RDLCK;
6088 	else
6089 		oldltype = F_UNLCK;
6090 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
6091 		/* nothing to do */
6092 		goto out;
6093 	error = nfsvno_advlock(vp, ltype, first, end, p);
6094 	if (error != 0) {
6095 		if (cfp != NULL) {
6096 			cfp->cl_clientid.lval[0] = 0;
6097 			cfp->cl_clientid.lval[1] = 0;
6098 			cfp->cl_first = 0;
6099 			cfp->cl_end = NFS64BITSSET;
6100 			cfp->cl_flags = NFSLCK_WRITE;
6101 			cfp->cl_ownerlen = 5;
6102 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
6103 		}
6104 		nfsrv_locallock_rollback(vp, lfp, p);
6105 	} else if (ltype != F_UNLCK) {
6106 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
6107 		    M_WAITOK);
6108 		rlp->rlck_first = first;
6109 		rlp->rlck_end = end;
6110 		rlp->rlck_type = oldltype;
6111 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
6112 	}
6113 
6114 out:
6115 	NFSEXITCODE(error);
6116 	return (error);
6117 }
6118 
6119 /*
6120  * Roll back local lock changes and free up the rollback list.
6121  */
6122 static void
6123 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
6124 {
6125 	struct nfsrollback *rlp, *nrlp;
6126 
6127 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
6128 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
6129 		    rlp->rlck_end, p);
6130 		free(rlp, M_NFSDROLLBACK);
6131 	}
6132 	LIST_INIT(&lfp->lf_rollback);
6133 }
6134 
6135 /*
6136  * Update local lock list and delete rollback list (ie now committed to the
6137  * local locks). Most of the work is done by the internal function.
6138  */
6139 static void
6140 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
6141     uint64_t end)
6142 {
6143 	struct nfsrollback *rlp, *nrlp;
6144 	struct nfslock *new_lop, *other_lop;
6145 
6146 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
6147 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
6148 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
6149 		    M_WAITOK);
6150 	else
6151 		other_lop = NULL;
6152 	new_lop->lo_flags = flags;
6153 	new_lop->lo_first = first;
6154 	new_lop->lo_end = end;
6155 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
6156 	if (new_lop != NULL)
6157 		free(new_lop, M_NFSDLOCK);
6158 	if (other_lop != NULL)
6159 		free(other_lop, M_NFSDLOCK);
6160 
6161 	/* and get rid of the rollback list */
6162 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
6163 		free(rlp, M_NFSDROLLBACK);
6164 	LIST_INIT(&lfp->lf_rollback);
6165 }
6166 
6167 /*
6168  * Lock the struct nfslockfile for local lock updating.
6169  */
6170 static void
6171 nfsrv_locklf(struct nfslockfile *lfp)
6172 {
6173 	int gotlock;
6174 
6175 	/* lf_usecount ensures *lfp won't be free'd */
6176 	lfp->lf_usecount++;
6177 	do {
6178 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
6179 		    NFSSTATEMUTEXPTR, NULL);
6180 	} while (gotlock == 0);
6181 	lfp->lf_usecount--;
6182 }
6183 
6184 /*
6185  * Unlock the struct nfslockfile after local lock updating.
6186  */
6187 static void
6188 nfsrv_unlocklf(struct nfslockfile *lfp)
6189 {
6190 
6191 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6192 }
6193 
6194 /*
6195  * Clear out all state for the NFSv4 server.
6196  * Must be called by a thread that can sleep when no nfsds are running.
6197  */
6198 void
6199 nfsrv_throwawayallstate(NFSPROC_T *p)
6200 {
6201 	struct nfsclient *clp, *nclp;
6202 	struct nfslockfile *lfp, *nlfp;
6203 	int i;
6204 
6205 	/*
6206 	 * For each client, clean out the state and then free the structure.
6207 	 */
6208 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6209 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
6210 			nfsrv_cleanclient(clp, p);
6211 			nfsrv_freedeleglist(&clp->lc_deleg);
6212 			nfsrv_freedeleglist(&clp->lc_olddeleg);
6213 			free(clp->lc_stateid, M_NFSDCLIENT);
6214 			free(clp, M_NFSDCLIENT);
6215 		}
6216 	}
6217 
6218 	/*
6219 	 * Also, free up any remaining lock file structures.
6220 	 */
6221 	for (i = 0; i < nfsrv_lockhashsize; i++) {
6222 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
6223 			printf("nfsd unload: fnd a lock file struct\n");
6224 			nfsrv_freenfslockfile(lfp);
6225 		}
6226 	}
6227 
6228 	/* And get rid of the deviceid structures and layouts. */
6229 	nfsrv_freealllayoutsanddevids();
6230 }
6231 
6232 /*
6233  * Check the sequence# for the session and slot provided as an argument.
6234  * Also, renew the lease if the session will return NFS_OK.
6235  */
6236 int
6237 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6238     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6239     uint32_t *sflagsp, NFSPROC_T *p)
6240 {
6241 	struct nfsdsession *sep;
6242 	struct nfssessionhash *shp;
6243 	int error;
6244 
6245 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6246 	NFSLOCKSESSION(shp);
6247 	sep = nfsrv_findsession(nd->nd_sessionid);
6248 	if (sep == NULL) {
6249 		NFSUNLOCKSESSION(shp);
6250 		return (NFSERR_BADSESSION);
6251 	}
6252 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6253 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6254 	if (error != 0) {
6255 		NFSUNLOCKSESSION(shp);
6256 		return (error);
6257 	}
6258 	if (cache_this != 0)
6259 		nd->nd_flag |= ND_SAVEREPLY;
6260 	/* Renew the lease. */
6261 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6262 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6263 	nd->nd_flag |= ND_IMPLIEDCLID;
6264 
6265 	/* Save maximum request and reply sizes. */
6266 	nd->nd_maxreq = sep->sess_maxreq;
6267 	nd->nd_maxresp = sep->sess_maxresp;
6268 
6269 	*sflagsp = 0;
6270 	if (sep->sess_clp->lc_req.nr_client == NULL ||
6271 	    (sep->sess_clp->lc_flags & LCL_CBDOWN) != 0)
6272 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
6273 	NFSUNLOCKSESSION(shp);
6274 	if (error == NFSERR_EXPIRED) {
6275 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6276 		error = 0;
6277 	} else if (error == NFSERR_ADMINREVOKED) {
6278 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6279 		error = 0;
6280 	}
6281 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6282 	return (0);
6283 }
6284 
6285 /*
6286  * Check/set reclaim complete for this session/clientid.
6287  */
6288 int
6289 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6290 {
6291 	struct nfsdsession *sep;
6292 	struct nfssessionhash *shp;
6293 	int error = 0;
6294 
6295 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6296 	NFSLOCKSTATE();
6297 	NFSLOCKSESSION(shp);
6298 	sep = nfsrv_findsession(nd->nd_sessionid);
6299 	if (sep == NULL) {
6300 		NFSUNLOCKSESSION(shp);
6301 		NFSUNLOCKSTATE();
6302 		return (NFSERR_BADSESSION);
6303 	}
6304 
6305 	if (onefs != 0)
6306 		sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6307 		/* Check to see if reclaim complete has already happened. */
6308 	else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6309 		error = NFSERR_COMPLETEALREADY;
6310 	else {
6311 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6312 		nfsrv_markreclaim(sep->sess_clp);
6313 	}
6314 	NFSUNLOCKSESSION(shp);
6315 	NFSUNLOCKSTATE();
6316 	return (error);
6317 }
6318 
6319 /*
6320  * Cache the reply in a session slot.
6321  */
6322 void
6323 nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m)
6324 {
6325 	struct nfsdsession *sep;
6326 	struct nfssessionhash *shp;
6327 	char *buf, *cp;
6328 #ifdef INET
6329 	struct sockaddr_in *sin;
6330 #endif
6331 #ifdef INET6
6332 	struct sockaddr_in6 *sin6;
6333 #endif
6334 
6335 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6336 	NFSLOCKSESSION(shp);
6337 	sep = nfsrv_findsession(nd->nd_sessionid);
6338 	if (sep == NULL) {
6339 		NFSUNLOCKSESSION(shp);
6340 		if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
6341 			buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK);
6342 			switch (nd->nd_nam->sa_family) {
6343 #ifdef INET
6344 			case AF_INET:
6345 				sin = (struct sockaddr_in *)nd->nd_nam;
6346 				cp = inet_ntop(sin->sin_family,
6347 				    &sin->sin_addr.s_addr, buf,
6348 				    INET6_ADDRSTRLEN);
6349 				break;
6350 #endif
6351 #ifdef INET6
6352 			case AF_INET6:
6353 				sin6 = (struct sockaddr_in6 *)nd->nd_nam;
6354 				cp = inet_ntop(sin6->sin6_family,
6355 				    &sin6->sin6_addr, buf, INET6_ADDRSTRLEN);
6356 				break;
6357 #endif
6358 			default:
6359 				cp = NULL;
6360 			}
6361 			if (cp != NULL)
6362 				printf("nfsrv_cache_session: no session "
6363 				    "IPaddr=%s\n", cp);
6364 			else
6365 				printf("nfsrv_cache_session: no session\n");
6366 			free(buf, M_TEMP);
6367 		}
6368 		m_freem(*m);
6369 		return;
6370 	}
6371 	nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat,
6372 	    m);
6373 	NFSUNLOCKSESSION(shp);
6374 }
6375 
6376 /*
6377  * Search for a session that matches the sessionid.
6378  */
6379 static struct nfsdsession *
6380 nfsrv_findsession(uint8_t *sessionid)
6381 {
6382 	struct nfsdsession *sep;
6383 	struct nfssessionhash *shp;
6384 
6385 	shp = NFSSESSIONHASH(sessionid);
6386 	LIST_FOREACH(sep, &shp->list, sess_hash) {
6387 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6388 			break;
6389 	}
6390 	return (sep);
6391 }
6392 
6393 /*
6394  * Destroy a session.
6395  */
6396 int
6397 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6398 {
6399 	int error, igotlock, samesess;
6400 
6401 	samesess = 0;
6402 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6403 	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6404 		samesess = 1;
6405 		if ((nd->nd_flag & ND_LASTOP) == 0)
6406 			return (NFSERR_BADSESSION);
6407 	}
6408 
6409 	/* Lock out other nfsd threads */
6410 	NFSLOCKV4ROOTMUTEX();
6411 	nfsv4_relref(&nfsv4rootfs_lock);
6412 	do {
6413 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6414 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6415 	} while (igotlock == 0);
6416 	NFSUNLOCKV4ROOTMUTEX();
6417 
6418 	error = nfsrv_freesession(NULL, sessionid);
6419 	if (error == 0 && samesess != 0)
6420 		nd->nd_flag &= ~ND_HASSEQUENCE;
6421 
6422 	NFSLOCKV4ROOTMUTEX();
6423 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6424 	NFSUNLOCKV4ROOTMUTEX();
6425 	return (error);
6426 }
6427 
6428 /*
6429  * Bind a connection to a session.
6430  * For now, only certain variants are supported, since the current session
6431  * structure can only handle a single backchannel entry, which will be
6432  * applied to all connections if it is set.
6433  */
6434 int
6435 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6436 {
6437 	struct nfssessionhash *shp;
6438 	struct nfsdsession *sep;
6439 	struct nfsclient *clp;
6440 	SVCXPRT *savxprt;
6441 	int error;
6442 
6443 	error = 0;
6444 	savxprt = NULL;
6445 	shp = NFSSESSIONHASH(sessionid);
6446 	NFSLOCKSTATE();
6447 	NFSLOCKSESSION(shp);
6448 	sep = nfsrv_findsession(sessionid);
6449 	if (sep != NULL) {
6450 		clp = sep->sess_clp;
6451 		if (*foreaftp == NFSCDFC4_BACK ||
6452 		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6453 		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6454 			/* Try to set up a backchannel. */
6455 			if (clp->lc_req.nr_client == NULL) {
6456 				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6457 				    "backchannel\n");
6458 				clp->lc_req.nr_client = (struct __rpc_client *)
6459 				    clnt_bck_create(nd->nd_xprt->xp_socket,
6460 				    sep->sess_cbprogram, NFSV4_CBVERS);
6461 			}
6462 			if (clp->lc_req.nr_client != NULL) {
6463 				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6464 				    "backchannel\n");
6465 				savxprt = sep->sess_cbsess.nfsess_xprt;
6466 				SVC_ACQUIRE(nd->nd_xprt);
6467 				CLNT_ACQUIRE(clp->lc_req.nr_client);
6468 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
6469 				/* Disable idle timeout. */
6470 				nd->nd_xprt->xp_idletimeout = 0;
6471 				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6472 				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6473 				clp->lc_flags |= LCL_DONEBINDCONN |
6474 				    LCL_NEEDSCBNULL;
6475 				clp->lc_flags &= ~LCL_CBDOWN;
6476 				if (*foreaftp == NFSCDFS4_BACK)
6477 					*foreaftp = NFSCDFS4_BACK;
6478 				else
6479 					*foreaftp = NFSCDFS4_BOTH;
6480 			} else if (*foreaftp != NFSCDFC4_BACK) {
6481 				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6482 				    "up backchannel\n");
6483 				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6484 				clp->lc_flags |= LCL_DONEBINDCONN;
6485 				*foreaftp = NFSCDFS4_FORE;
6486 			} else {
6487 				error = NFSERR_NOTSUPP;
6488 				printf("nfsrv_bindconnsess: Can't add "
6489 				    "backchannel\n");
6490 			}
6491 		} else {
6492 			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6493 			clp->lc_flags |= LCL_DONEBINDCONN;
6494 			*foreaftp = NFSCDFS4_FORE;
6495 		}
6496 	} else
6497 		error = NFSERR_BADSESSION;
6498 	NFSUNLOCKSESSION(shp);
6499 	NFSUNLOCKSTATE();
6500 	if (savxprt != NULL)
6501 		SVC_RELEASE(savxprt);
6502 	return (error);
6503 }
6504 
6505 /*
6506  * Free up a session structure.
6507  */
6508 static int
6509 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
6510 {
6511 	struct nfssessionhash *shp;
6512 	int i;
6513 
6514 	NFSLOCKSTATE();
6515 	if (sep == NULL) {
6516 		shp = NFSSESSIONHASH(sessionid);
6517 		NFSLOCKSESSION(shp);
6518 		sep = nfsrv_findsession(sessionid);
6519 	} else {
6520 		shp = NFSSESSIONHASH(sep->sess_sessionid);
6521 		NFSLOCKSESSION(shp);
6522 	}
6523 	if (sep != NULL) {
6524 		sep->sess_refcnt--;
6525 		if (sep->sess_refcnt > 0) {
6526 			NFSUNLOCKSESSION(shp);
6527 			NFSUNLOCKSTATE();
6528 			return (NFSERR_BACKCHANBUSY);
6529 		}
6530 		LIST_REMOVE(sep, sess_hash);
6531 		LIST_REMOVE(sep, sess_list);
6532 	}
6533 	NFSUNLOCKSESSION(shp);
6534 	NFSUNLOCKSTATE();
6535 	if (sep == NULL)
6536 		return (NFSERR_BADSESSION);
6537 	for (i = 0; i < NFSV4_SLOTS; i++)
6538 		if (sep->sess_slots[i].nfssl_reply != NULL)
6539 			m_freem(sep->sess_slots[i].nfssl_reply);
6540 	if (sep->sess_cbsess.nfsess_xprt != NULL)
6541 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6542 	free(sep, M_NFSDSESSION);
6543 	return (0);
6544 }
6545 
6546 /*
6547  * Free a stateid.
6548  * RFC5661 says that it should fail when there are associated opens, locks
6549  * or delegations. Since stateids represent opens, I don't see how you can
6550  * free an open stateid (it will be free'd when closed), so this function
6551  * only works for lock stateids (freeing the lock_owner) or delegations.
6552  */
6553 int
6554 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6555     NFSPROC_T *p)
6556 {
6557 	struct nfsclient *clp;
6558 	struct nfsstate *stp;
6559 	int error;
6560 
6561 	NFSLOCKSTATE();
6562 	/*
6563 	 * Look up the stateid
6564 	 */
6565 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6566 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6567 	if (error == 0) {
6568 		/* First, check for a delegation. */
6569 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6570 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6571 			    NFSX_STATEIDOTHER))
6572 				break;
6573 		}
6574 		if (stp != NULL) {
6575 			nfsrv_freedeleg(stp);
6576 			NFSUNLOCKSTATE();
6577 			return (error);
6578 		}
6579 	}
6580 	/* Not a delegation, try for a lock_owner. */
6581 	if (error == 0)
6582 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6583 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6584 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6585 		/* Not a lock_owner stateid. */
6586 		error = NFSERR_LOCKSHELD;
6587 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6588 		error = NFSERR_LOCKSHELD;
6589 	if (error == 0)
6590 		nfsrv_freelockowner(stp, NULL, 0, p);
6591 	NFSUNLOCKSTATE();
6592 	return (error);
6593 }
6594 
6595 /*
6596  * Test a stateid.
6597  */
6598 int
6599 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6600     NFSPROC_T *p)
6601 {
6602 	struct nfsclient *clp;
6603 	struct nfsstate *stp;
6604 	int error;
6605 
6606 	NFSLOCKSTATE();
6607 	/*
6608 	 * Look up the stateid
6609 	 */
6610 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6611 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6612 	if (error == 0)
6613 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6614 	if (error == 0 && stateidp->seqid != 0 &&
6615 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6616 		error = NFSERR_OLDSTATEID;
6617 	NFSUNLOCKSTATE();
6618 	return (error);
6619 }
6620 
6621 /*
6622  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6623  */
6624 static int
6625 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6626     int dont_replycache, struct nfsdsession **sepp, int *slotposp)
6627 {
6628 	struct nfsdsession *sep;
6629 	uint32_t *tl, slotseq = 0;
6630 	int maxslot;
6631 	uint8_t sessionid[NFSX_V4SESSIONID];
6632 	int error;
6633 
6634 	error = nfsv4_getcbsession(clp, sepp);
6635 	if (error != 0)
6636 		return (error);
6637 	sep = *sepp;
6638 	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, slotposp, &maxslot,
6639 	    &slotseq, sessionid);
6640 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6641 
6642 	/* Build the Sequence arguments. */
6643 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6644 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6645 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6646 	nd->nd_slotseq = tl;
6647 	*tl++ = txdr_unsigned(slotseq);
6648 	*tl++ = txdr_unsigned(*slotposp);
6649 	*tl++ = txdr_unsigned(maxslot);
6650 	if (dont_replycache == 0)
6651 		*tl++ = newnfs_true;
6652 	else
6653 		*tl++ = newnfs_false;
6654 	*tl = 0;			/* No referring call list, for now. */
6655 	nd->nd_flag |= ND_HASSEQUENCE;
6656 	return (0);
6657 }
6658 
6659 /*
6660  * Get a session for the callback.
6661  */
6662 static int
6663 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6664 {
6665 	struct nfsdsession *sep;
6666 
6667 	NFSLOCKSTATE();
6668 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6669 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6670 			break;
6671 	}
6672 	if (sep == NULL) {
6673 		NFSUNLOCKSTATE();
6674 		return (NFSERR_BADSESSION);
6675 	}
6676 	sep->sess_refcnt++;
6677 	*sepp = sep;
6678 	NFSUNLOCKSTATE();
6679 	return (0);
6680 }
6681 
6682 /*
6683  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6684  * exit, since those transports will all be going away.
6685  * This is only called after all the nfsd threads are done performing RPCs,
6686  * so locking shouldn't be an issue.
6687  */
6688 void
6689 nfsrv_freeallbackchannel_xprts(void)
6690 {
6691 	struct nfsdsession *sep;
6692 	struct nfsclient *clp;
6693 	SVCXPRT *xprt;
6694 	int i;
6695 
6696 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6697 		LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6698 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6699 				xprt = sep->sess_cbsess.nfsess_xprt;
6700 				sep->sess_cbsess.nfsess_xprt = NULL;
6701 				if (xprt != NULL)
6702 					SVC_RELEASE(xprt);
6703 			}
6704 		}
6705 	}
6706 }
6707 
6708 /*
6709  * Do a layout commit.  Actually just call nfsrv_updatemdsattr().
6710  * I have no idea if the rest of these arguments will ever be useful?
6711  */
6712 int
6713 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6714     int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6715     int hasnewmtime, struct timespec *newmtimep, int reclaim,
6716     nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6717     uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6718 {
6719 	struct nfsvattr na;
6720 	int error;
6721 
6722 	error = nfsrv_updatemdsattr(vp, &na, p);
6723 	if (error == 0) {
6724 		*hasnewsizep = 1;
6725 		*newsizep = na.na_size;
6726 	}
6727 	return (error);
6728 }
6729 
6730 /*
6731  * Try and get a layout.
6732  */
6733 int
6734 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6735     int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6736     uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6737     int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6738 {
6739 	struct nfslayouthash *lhyp;
6740 	struct nfslayout *lyp;
6741 	char *devid;
6742 	fhandle_t fh, *dsfhp;
6743 	int error, mirrorcnt;
6744 
6745 	if (nfsrv_devidcnt == 0)
6746 		return (NFSERR_UNKNLAYOUTTYPE);
6747 
6748 	if (*offset != 0)
6749 		printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6750 		    (uintmax_t)*len);
6751 	error = nfsvno_getfh(vp, &fh, p);
6752 	NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6753 	if (error != 0)
6754 		return (error);
6755 
6756 	/*
6757 	 * For now, all layouts are for entire files.
6758 	 * Only issue Read/Write layouts if requested for a non-readonly fs.
6759 	 */
6760 	if (NFSVNO_EXRDONLY(exp)) {
6761 		if (*iomode == NFSLAYOUTIOMODE_RW)
6762 			return (NFSERR_LAYOUTTRYLATER);
6763 		*iomode = NFSLAYOUTIOMODE_READ;
6764 	}
6765 	if (*iomode != NFSLAYOUTIOMODE_RW)
6766 		*iomode = NFSLAYOUTIOMODE_READ;
6767 
6768 	/*
6769 	 * Check to see if a write layout can be issued for this file.
6770 	 * This is used during mirror recovery to avoid RW layouts being
6771 	 * issued for a file while it is being copied to the recovered
6772 	 * mirror.
6773 	 */
6774 	if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6775 		return (NFSERR_LAYOUTTRYLATER);
6776 
6777 	*retonclose = 0;
6778 	*offset = 0;
6779 	*len = UINT64_MAX;
6780 
6781 	/* First, see if a layout already exists and return if found. */
6782 	lhyp = NFSLAYOUTHASH(&fh);
6783 	NFSLOCKLAYOUT(lhyp);
6784 	error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6785 	NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6786 	/*
6787 	 * Not sure if the seqid must be the same, so I won't check it.
6788 	 */
6789 	if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6790 	    stateidp->other[1] != lyp->lay_stateid.other[1] ||
6791 	    stateidp->other[2] != lyp->lay_stateid.other[2])) {
6792 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6793 			NFSUNLOCKLAYOUT(lhyp);
6794 			NFSD_DEBUG(1, "ret bad stateid\n");
6795 			return (NFSERR_BADSTATEID);
6796 		}
6797 		/*
6798 		 * I believe we get here because there is a race between
6799 		 * the client processing the CBLAYOUTRECALL and the layout
6800 		 * being deleted here on the server.
6801 		 * The client has now done a LayoutGet with a non-layout
6802 		 * stateid, as it would when there is no layout.
6803 		 * As such, free this layout and set error == NFSERR_BADSTATEID
6804 		 * so the code below will create a new layout structure as
6805 		 * would happen if no layout was found.
6806 		 * "lyp" will be set before being used below, but set it NULL
6807 		 * as a safety belt.
6808 		 */
6809 		nfsrv_freelayout(&lhyp->list, lyp);
6810 		lyp = NULL;
6811 		error = NFSERR_BADSTATEID;
6812 	}
6813 	if (error == 0) {
6814 		if (lyp->lay_layoutlen > maxcnt) {
6815 			NFSUNLOCKLAYOUT(lhyp);
6816 			NFSD_DEBUG(1, "ret layout too small\n");
6817 			return (NFSERR_TOOSMALL);
6818 		}
6819 		if (*iomode == NFSLAYOUTIOMODE_RW)
6820 			lyp->lay_flags |= NFSLAY_RW;
6821 		else
6822 			lyp->lay_flags |= NFSLAY_READ;
6823 		NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6824 		*layoutlenp = lyp->lay_layoutlen;
6825 		if (++lyp->lay_stateid.seqid == 0)
6826 			lyp->lay_stateid.seqid = 1;
6827 		stateidp->seqid = lyp->lay_stateid.seqid;
6828 		NFSUNLOCKLAYOUT(lhyp);
6829 		NFSD_DEBUG(4, "ret fnd layout\n");
6830 		return (0);
6831 	}
6832 	NFSUNLOCKLAYOUT(lhyp);
6833 
6834 	/* Find the device id and file handle. */
6835 	dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6836 	devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6837 	error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid);
6838 	NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6839 	if (error == 0) {
6840 		if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6841 			if (NFSX_V4FILELAYOUT > maxcnt)
6842 				error = NFSERR_TOOSMALL;
6843 			else
6844 				lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6845 				    devid, vp->v_mount->mnt_stat.f_fsid);
6846 		} else {
6847 			if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt)
6848 				error = NFSERR_TOOSMALL;
6849 			else
6850 				lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6851 				    &fh, dsfhp, devid,
6852 				    vp->v_mount->mnt_stat.f_fsid);
6853 		}
6854 	}
6855 	free(dsfhp, M_TEMP);
6856 	free(devid, M_TEMP);
6857 	if (error != 0)
6858 		return (error);
6859 
6860 	/*
6861 	 * Now, add this layout to the list.
6862 	 */
6863 	error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6864 	NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6865 	/*
6866 	 * The lyp will be set to NULL by nfsrv_addlayout() if it
6867 	 * linked the new structure into the lists.
6868 	 */
6869 	free(lyp, M_NFSDSTATE);
6870 	return (error);
6871 }
6872 
6873 /*
6874  * Generate a File Layout.
6875  */
6876 static struct nfslayout *
6877 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6878     fhandle_t *dsfhp, char *devid, fsid_t fs)
6879 {
6880 	uint32_t *tl;
6881 	struct nfslayout *lyp;
6882 	uint64_t pattern_offset;
6883 
6884 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6885 	    M_WAITOK | M_ZERO);
6886 	lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6887 	if (iomode == NFSLAYOUTIOMODE_RW)
6888 		lyp->lay_flags = NFSLAY_RW;
6889 	else
6890 		lyp->lay_flags = NFSLAY_READ;
6891 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6892 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6893 	lyp->lay_fsid = fs;
6894 
6895 	/* Fill in the xdr for the files layout. */
6896 	tl = (uint32_t *)lyp->lay_xdr;
6897 	NFSBCOPY(devid, tl, NFSX_V4DEVICEID);		/* Device ID. */
6898 	tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6899 
6900 	/* Set the stripe size to the maximum I/O size. */
6901 	*tl++ = txdr_unsigned(NFS_SRVMAXIO & NFSFLAYUTIL_STRIPE_MASK);
6902 	*tl++ = 0;					/* 1st stripe index. */
6903 	pattern_offset = 0;
6904 	txdr_hyper(pattern_offset, tl); tl += 2;	/* Pattern offset. */
6905 	*tl++ = txdr_unsigned(1);			/* 1 file handle. */
6906 	*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6907 	NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6908 	lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
6909 	return (lyp);
6910 }
6911 
6912 #define	FLEX_OWNERID	"999"
6913 #define	FLEX_UID0	"0"
6914 /*
6915  * Generate a Flex File Layout.
6916  * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
6917  * string goes on the wire, it isn't supposed to be used by the client,
6918  * since this server uses tight coupling.
6919  * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
6920  * a string of "0". This works around the Linux Flex File Layout driver bug
6921  * which uses the synthetic uid/gid strings for the "tightly coupled" case.
6922  */
6923 static struct nfslayout *
6924 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
6925     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs)
6926 {
6927 	uint32_t *tl;
6928 	struct nfslayout *lyp;
6929 	uint64_t lenval;
6930 	int i;
6931 
6932 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt),
6933 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
6934 	lyp->lay_type = NFSLAYOUT_FLEXFILE;
6935 	if (iomode == NFSLAYOUTIOMODE_RW)
6936 		lyp->lay_flags = NFSLAY_RW;
6937 	else
6938 		lyp->lay_flags = NFSLAY_READ;
6939 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6940 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6941 	lyp->lay_fsid = fs;
6942 	lyp->lay_mirrorcnt = mirrorcnt;
6943 
6944 	/* Fill in the xdr for the files layout. */
6945 	tl = (uint32_t *)lyp->lay_xdr;
6946 	lenval = 0;
6947 	txdr_hyper(lenval, tl); tl += 2;		/* Stripe unit. */
6948 	*tl++ = txdr_unsigned(mirrorcnt);		/* # of mirrors. */
6949 	for (i = 0; i < mirrorcnt; i++) {
6950 		*tl++ = txdr_unsigned(1);		/* One stripe. */
6951 		NFSBCOPY(devid, tl, NFSX_V4DEVICEID);	/* Device ID. */
6952 		tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6953 		devid += NFSX_V4DEVICEID;
6954 		*tl++ = txdr_unsigned(1);		/* Efficiency. */
6955 		*tl++ = 0;				/* Proxy Stateid. */
6956 		*tl++ = 0x55555555;
6957 		*tl++ = 0x55555555;
6958 		*tl++ = 0x55555555;
6959 		*tl++ = txdr_unsigned(1);		/* 1 file handle. */
6960 		*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6961 		NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6962 		tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
6963 		dsfhp++;
6964 		if (nfsrv_flexlinuxhack != 0) {
6965 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
6966 			*tl = 0;		/* 0 pad string. */
6967 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6968 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
6969 			*tl = 0;		/* 0 pad string. */
6970 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6971 		} else {
6972 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6973 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6974 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6975 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6976 		}
6977 	}
6978 	*tl++ = txdr_unsigned(0);		/* ff_flags. */
6979 	*tl = txdr_unsigned(60);		/* Status interval hint. */
6980 	lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt);
6981 	return (lyp);
6982 }
6983 
6984 /*
6985  * Parse and process Flex File errors returned via LayoutReturn.
6986  */
6987 static void
6988 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
6989     NFSPROC_T *p)
6990 {
6991 	uint32_t *tl;
6992 	int cnt, errcnt, i, j, opnum, stat;
6993 	char devid[NFSX_V4DEVICEID];
6994 
6995 	tl = layp;
6996 	cnt = fxdr_unsigned(int, *tl++);
6997 	NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
6998 	for (i = 0; i < cnt; i++) {
6999 		/* Skip offset, length and stateid for now. */
7000 		tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
7001 		errcnt = fxdr_unsigned(int, *tl++);
7002 		NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
7003 		for (j = 0; j < errcnt; j++) {
7004 			NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
7005 			tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7006 			stat = fxdr_unsigned(int, *tl++);
7007 			opnum = fxdr_unsigned(int, *tl++);
7008 			NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
7009 			    stat);
7010 			/*
7011 			 * Except for NFSERR_ACCES and NFSERR_STALE errors,
7012 			 * disable the mirror.
7013 			 */
7014 			if (stat != NFSERR_ACCES && stat != NFSERR_STALE)
7015 				nfsrv_delds(devid, p);
7016 		}
7017 	}
7018 }
7019 
7020 /*
7021  * This function removes all flex file layouts which has a mirror with
7022  * a device id that matches the argument.
7023  * Called when the DS represented by the device id has failed.
7024  */
7025 void
7026 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
7027 {
7028 	uint32_t *tl;
7029 	struct nfslayout *lyp, *nlyp;
7030 	struct nfslayouthash *lhyp;
7031 	struct nfslayouthead loclyp;
7032 	int i, j;
7033 
7034 	NFSD_DEBUG(4, "flexmirrordel\n");
7035 	/* Move all layouts found onto a local list. */
7036 	TAILQ_INIT(&loclyp);
7037 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7038 		lhyp = &nfslayouthash[i];
7039 		NFSLOCKLAYOUT(lhyp);
7040 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7041 			if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
7042 			    lyp->lay_mirrorcnt > 1) {
7043 				NFSD_DEBUG(4, "possible match\n");
7044 				tl = lyp->lay_xdr;
7045 				tl += 3;
7046 				for (j = 0; j < lyp->lay_mirrorcnt; j++) {
7047 					tl++;
7048 					if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
7049 					    == 0) {
7050 						/* Found one. */
7051 						NFSD_DEBUG(4, "fnd one\n");
7052 						TAILQ_REMOVE(&lhyp->list, lyp,
7053 						    lay_list);
7054 						TAILQ_INSERT_HEAD(&loclyp, lyp,
7055 						    lay_list);
7056 						break;
7057 					}
7058 					tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
7059 					    NFSM_RNDUP(NFSX_V4PNFSFH) /
7060 					    NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
7061 				}
7062 			}
7063 		}
7064 		NFSUNLOCKLAYOUT(lhyp);
7065 	}
7066 
7067 	/* Now, try to do a Layout recall for each one found. */
7068 	TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
7069 		NFSD_DEBUG(4, "do layout recall\n");
7070 		/*
7071 		 * The layout stateid.seqid needs to be incremented
7072 		 * before doing a LAYOUT_RECALL callback.
7073 		 */
7074 		if (++lyp->lay_stateid.seqid == 0)
7075 			lyp->lay_stateid.seqid = 1;
7076 		nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
7077 		    &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
7078 		nfsrv_freelayout(&loclyp, lyp);
7079 	}
7080 }
7081 
7082 /*
7083  * Do a recall callback to the client for this layout.
7084  */
7085 static int
7086 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
7087     struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
7088 {
7089 	struct nfsclient *clp;
7090 	int error;
7091 
7092 	NFSD_DEBUG(4, "nfsrv_recalllayout\n");
7093 	error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
7094 	    0, NULL, p);
7095 	NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
7096 	if (error != 0) {
7097 		printf("nfsrv_recalllayout: getclient err=%d\n", error);
7098 		return (error);
7099 	}
7100 	if ((clp->lc_flags & LCL_NFSV41) != 0) {
7101 		error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
7102 		    stateidp, changed, fhp, NULL, NULL, laytype, p);
7103 		/* If lyp != NULL, handle an error return here. */
7104 		if (error != 0 && lyp != NULL) {
7105 			NFSDRECALLLOCK();
7106 			/*
7107 			 * Mark it returned, since no layout recall
7108 			 * has been done.
7109 			 * All errors seem to be non-recoverable, although
7110 			 * NFSERR_NOMATCHLAYOUT is a normal event.
7111 			 */
7112 			if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
7113 				lyp->lay_flags |= NFSLAY_RETURNED;
7114 				wakeup(lyp);
7115 			}
7116 			NFSDRECALLUNLOCK();
7117 			if (error != NFSERR_NOMATCHLAYOUT)
7118 				printf("nfsrv_recalllayout: err=%d\n", error);
7119 		}
7120 	} else
7121 		printf("nfsrv_recalllayout: clp not NFSv4.1\n");
7122 	return (error);
7123 }
7124 
7125 /*
7126  * Find a layout to recall when we exceed our high water mark.
7127  */
7128 void
7129 nfsrv_recalloldlayout(NFSPROC_T *p)
7130 {
7131 	struct nfslayouthash *lhyp;
7132 	struct nfslayout *lyp;
7133 	nfsquad_t clientid;
7134 	nfsv4stateid_t stateid;
7135 	fhandle_t fh;
7136 	int error, laytype = 0, ret;
7137 
7138 	lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
7139 	NFSLOCKLAYOUT(lhyp);
7140 	TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
7141 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
7142 			lyp->lay_flags |= NFSLAY_CALLB;
7143 			/*
7144 			 * The layout stateid.seqid needs to be incremented
7145 			 * before doing a LAYOUT_RECALL callback.
7146 			 */
7147 			if (++lyp->lay_stateid.seqid == 0)
7148 				lyp->lay_stateid.seqid = 1;
7149 			clientid = lyp->lay_clientid;
7150 			stateid = lyp->lay_stateid;
7151 			NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
7152 			laytype = lyp->lay_type;
7153 			break;
7154 		}
7155 	}
7156 	NFSUNLOCKLAYOUT(lhyp);
7157 	if (lyp != NULL) {
7158 		error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
7159 		    laytype, p);
7160 		if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
7161 			NFSD_DEBUG(4, "recallold=%d\n", error);
7162 		if (error != 0) {
7163 			NFSLOCKLAYOUT(lhyp);
7164 			/*
7165 			 * Since the hash list was unlocked, we need to
7166 			 * find it again.
7167 			 */
7168 			ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
7169 			    &lyp);
7170 			if (ret == 0 &&
7171 			    (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
7172 			    lyp->lay_stateid.other[0] == stateid.other[0] &&
7173 			    lyp->lay_stateid.other[1] == stateid.other[1] &&
7174 			    lyp->lay_stateid.other[2] == stateid.other[2]) {
7175 				/*
7176 				 * The client no longer knows this layout, so
7177 				 * it can be free'd now.
7178 				 */
7179 				if (error == NFSERR_NOMATCHLAYOUT)
7180 					nfsrv_freelayout(&lhyp->list, lyp);
7181 				else {
7182 					/*
7183 					 * Leave it to be tried later by
7184 					 * clearing NFSLAY_CALLB and moving
7185 					 * it to the head of the list, so it
7186 					 * won't be tried again for a while.
7187 					 */
7188 					lyp->lay_flags &= ~NFSLAY_CALLB;
7189 					TAILQ_REMOVE(&lhyp->list, lyp,
7190 					    lay_list);
7191 					TAILQ_INSERT_HEAD(&lhyp->list, lyp,
7192 					    lay_list);
7193 				}
7194 			}
7195 			NFSUNLOCKLAYOUT(lhyp);
7196 		}
7197 	}
7198 }
7199 
7200 /*
7201  * Try and return layout(s).
7202  */
7203 int
7204 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
7205     int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
7206     int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
7207     struct ucred *cred, NFSPROC_T *p)
7208 {
7209 	struct nfsvattr na;
7210 	struct nfslayouthash *lhyp;
7211 	struct nfslayout *lyp;
7212 	fhandle_t fh;
7213 	int error = 0;
7214 
7215 	*fndp = 0;
7216 	if (kind == NFSV4LAYOUTRET_FILE) {
7217 		error = nfsvno_getfh(vp, &fh, p);
7218 		if (error == 0) {
7219 			error = nfsrv_updatemdsattr(vp, &na, p);
7220 			if (error != 0)
7221 				printf("nfsrv_layoutreturn: updatemdsattr"
7222 				    " failed=%d\n", error);
7223 		}
7224 		if (error == 0) {
7225 			if (reclaim == newnfs_true) {
7226 				error = nfsrv_checkgrace(NULL, NULL,
7227 				    NFSLCK_RECLAIM);
7228 				if (error != NFSERR_NOGRACE)
7229 					error = 0;
7230 				return (error);
7231 			}
7232 			lhyp = NFSLAYOUTHASH(&fh);
7233 			NFSDRECALLLOCK();
7234 			NFSLOCKLAYOUT(lhyp);
7235 			error = nfsrv_findlayout(&nd->nd_clientid, &fh,
7236 			    layouttype, p, &lyp);
7237 			NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
7238 			if (error == 0 &&
7239 			    stateidp->other[0] == lyp->lay_stateid.other[0] &&
7240 			    stateidp->other[1] == lyp->lay_stateid.other[1] &&
7241 			    stateidp->other[2] == lyp->lay_stateid.other[2]) {
7242 				NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
7243 				    " %x %x %x laystateid %d %x %x %x"
7244 				    " off=%ju len=%ju flgs=0x%x\n",
7245 				    stateidp->seqid, stateidp->other[0],
7246 				    stateidp->other[1], stateidp->other[2],
7247 				    lyp->lay_stateid.seqid,
7248 				    lyp->lay_stateid.other[0],
7249 				    lyp->lay_stateid.other[1],
7250 				    lyp->lay_stateid.other[2],
7251 				    (uintmax_t)offset, (uintmax_t)len,
7252 				    lyp->lay_flags);
7253 				if (++lyp->lay_stateid.seqid == 0)
7254 					lyp->lay_stateid.seqid = 1;
7255 				stateidp->seqid = lyp->lay_stateid.seqid;
7256 				if (offset == 0 && len == UINT64_MAX) {
7257 					if ((iomode & NFSLAYOUTIOMODE_READ) !=
7258 					    0)
7259 						lyp->lay_flags &= ~NFSLAY_READ;
7260 					if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7261 						lyp->lay_flags &= ~NFSLAY_RW;
7262 					if ((lyp->lay_flags & (NFSLAY_READ |
7263 					    NFSLAY_RW)) == 0)
7264 						nfsrv_freelayout(&lhyp->list,
7265 						    lyp);
7266 					else
7267 						*fndp = 1;
7268 				} else
7269 					*fndp = 1;
7270 			}
7271 			NFSUNLOCKLAYOUT(lhyp);
7272 			/* Search the nfsrv_recalllist for a match. */
7273 			TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7274 				if (NFSBCMP(&lyp->lay_fh, &fh,
7275 				    sizeof(fh)) == 0 &&
7276 				    lyp->lay_clientid.qval ==
7277 				    nd->nd_clientid.qval &&
7278 				    stateidp->other[0] ==
7279 				    lyp->lay_stateid.other[0] &&
7280 				    stateidp->other[1] ==
7281 				    lyp->lay_stateid.other[1] &&
7282 				    stateidp->other[2] ==
7283 				    lyp->lay_stateid.other[2]) {
7284 					lyp->lay_flags |= NFSLAY_RETURNED;
7285 					wakeup(lyp);
7286 					error = 0;
7287 				}
7288 			}
7289 			NFSDRECALLUNLOCK();
7290 		}
7291 		if (layouttype == NFSLAYOUT_FLEXFILE)
7292 			nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7293 	} else if (kind == NFSV4LAYOUTRET_FSID)
7294 		nfsrv_freelayouts(&nd->nd_clientid,
7295 		    &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7296 	else if (kind == NFSV4LAYOUTRET_ALL)
7297 		nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7298 	else
7299 		error = NFSERR_INVAL;
7300 	if (error == -1)
7301 		error = 0;
7302 	return (error);
7303 }
7304 
7305 /*
7306  * Look for an existing layout.
7307  */
7308 static int
7309 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7310     NFSPROC_T *p, struct nfslayout **lypp)
7311 {
7312 	struct nfslayouthash *lhyp;
7313 	struct nfslayout *lyp;
7314 	int ret;
7315 
7316 	*lypp = NULL;
7317 	ret = 0;
7318 	lhyp = NFSLAYOUTHASH(fhp);
7319 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7320 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7321 		    lyp->lay_clientid.qval == clientidp->qval &&
7322 		    lyp->lay_type == laytype)
7323 			break;
7324 	}
7325 	if (lyp != NULL)
7326 		*lypp = lyp;
7327 	else
7328 		ret = -1;
7329 	return (ret);
7330 }
7331 
7332 /*
7333  * Add the new layout, as required.
7334  */
7335 static int
7336 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7337     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7338 {
7339 	struct nfsclient *clp;
7340 	struct nfslayouthash *lhyp;
7341 	struct nfslayout *lyp, *nlyp;
7342 	fhandle_t *fhp;
7343 	int error;
7344 
7345 	KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7346 	    ("nfsrv_layoutget: no nd_clientid\n"));
7347 	lyp = *lypp;
7348 	fhp = &lyp->lay_fh;
7349 	NFSLOCKSTATE();
7350 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7351 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7352 	if (error != 0) {
7353 		NFSUNLOCKSTATE();
7354 		return (error);
7355 	}
7356 	lyp->lay_stateid.seqid = stateidp->seqid = 1;
7357 	lyp->lay_stateid.other[0] = stateidp->other[0] =
7358 	    clp->lc_clientid.lval[0];
7359 	lyp->lay_stateid.other[1] = stateidp->other[1] =
7360 	    clp->lc_clientid.lval[1];
7361 	lyp->lay_stateid.other[2] = stateidp->other[2] =
7362 	    nfsrv_nextstateindex(clp);
7363 	NFSUNLOCKSTATE();
7364 
7365 	lhyp = NFSLAYOUTHASH(fhp);
7366 	NFSLOCKLAYOUT(lhyp);
7367 	TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7368 		if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7369 		    nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7370 			break;
7371 	}
7372 	if (nlyp != NULL) {
7373 		/* A layout already exists, so use it. */
7374 		nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7375 		NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7376 		*layoutlenp = nlyp->lay_layoutlen;
7377 		if (++nlyp->lay_stateid.seqid == 0)
7378 			nlyp->lay_stateid.seqid = 1;
7379 		stateidp->seqid = nlyp->lay_stateid.seqid;
7380 		stateidp->other[0] = nlyp->lay_stateid.other[0];
7381 		stateidp->other[1] = nlyp->lay_stateid.other[1];
7382 		stateidp->other[2] = nlyp->lay_stateid.other[2];
7383 		NFSUNLOCKLAYOUT(lhyp);
7384 		return (0);
7385 	}
7386 
7387 	/* Insert the new layout in the lists. */
7388 	*lypp = NULL;
7389 	atomic_add_int(&nfsrv_layoutcnt, 1);
7390 	NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7391 	*layoutlenp = lyp->lay_layoutlen;
7392 	TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7393 	NFSUNLOCKLAYOUT(lhyp);
7394 	return (0);
7395 }
7396 
7397 /*
7398  * Get the devinfo for a deviceid.
7399  */
7400 int
7401 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7402     uint32_t *notify, int *devaddrlen, char **devaddr)
7403 {
7404 	struct nfsdevice *ds;
7405 
7406 	if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7407 	     NFSLAYOUT_FLEXFILE) ||
7408 	    (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7409 		return (NFSERR_UNKNLAYOUTTYPE);
7410 
7411 	/*
7412 	 * Now, search for the device id.  Note that the structures won't go
7413 	 * away, but the order changes in the list.  As such, the lock only
7414 	 * needs to be held during the search through the list.
7415 	 */
7416 	NFSDDSLOCK();
7417 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7418 		if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7419 		    ds->nfsdev_nmp != NULL)
7420 			break;
7421 	}
7422 	NFSDDSUNLOCK();
7423 	if (ds == NULL)
7424 		return (NFSERR_NOENT);
7425 
7426 	/* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7427 	*devaddrlen = 0;
7428 	if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7429 		*devaddrlen = ds->nfsdev_fileaddrlen;
7430 		*devaddr = ds->nfsdev_fileaddr;
7431 	} else if (layouttype == NFSLAYOUT_FLEXFILE) {
7432 		*devaddrlen = ds->nfsdev_flexaddrlen;
7433 		*devaddr = ds->nfsdev_flexaddr;
7434 	}
7435 	if (*devaddrlen == 0)
7436 		return (NFSERR_UNKNLAYOUTTYPE);
7437 
7438 	/*
7439 	 * The XDR overhead is 3 unsigned values: layout_type,
7440 	 * length_of_address and notify bitmap.
7441 	 * If the notify array is changed to not all zeros, the
7442 	 * count of unsigned values must be increased.
7443 	 */
7444 	if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7445 	    3 * NFSX_UNSIGNED) {
7446 		*maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7447 		return (NFSERR_TOOSMALL);
7448 	}
7449 	return (0);
7450 }
7451 
7452 /*
7453  * Free a list of layout state structures.
7454  */
7455 static void
7456 nfsrv_freelayoutlist(nfsquad_t clientid)
7457 {
7458 	struct nfslayouthash *lhyp;
7459 	struct nfslayout *lyp, *nlyp;
7460 	int i;
7461 
7462 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7463 		lhyp = &nfslayouthash[i];
7464 		NFSLOCKLAYOUT(lhyp);
7465 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7466 			if (lyp->lay_clientid.qval == clientid.qval)
7467 				nfsrv_freelayout(&lhyp->list, lyp);
7468 		}
7469 		NFSUNLOCKLAYOUT(lhyp);
7470 	}
7471 }
7472 
7473 /*
7474  * Free up a layout.
7475  */
7476 static void
7477 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7478 {
7479 
7480 	NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7481 	atomic_add_int(&nfsrv_layoutcnt, -1);
7482 	TAILQ_REMOVE(lhp, lyp, lay_list);
7483 	free(lyp, M_NFSDSTATE);
7484 }
7485 
7486 /*
7487  * Free up a device id.
7488  */
7489 void
7490 nfsrv_freeonedevid(struct nfsdevice *ds)
7491 {
7492 	int i;
7493 
7494 	atomic_add_int(&nfsrv_devidcnt, -1);
7495 	vrele(ds->nfsdev_dvp);
7496 	for (i = 0; i < nfsrv_dsdirsize; i++)
7497 		if (ds->nfsdev_dsdir[i] != NULL)
7498 			vrele(ds->nfsdev_dsdir[i]);
7499 	free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7500 	free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7501 	free(ds->nfsdev_host, M_NFSDSTATE);
7502 	free(ds, M_NFSDSTATE);
7503 }
7504 
7505 /*
7506  * Free up a device id and its mirrors.
7507  */
7508 static void
7509 nfsrv_freedevid(struct nfsdevice *ds)
7510 {
7511 
7512 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7513 	nfsrv_freeonedevid(ds);
7514 }
7515 
7516 /*
7517  * Free all layouts and device ids.
7518  * Done when the nfsd threads are shut down since there may be a new
7519  * modified device id list created when the nfsd is restarted.
7520  */
7521 void
7522 nfsrv_freealllayoutsanddevids(void)
7523 {
7524 	struct nfsdontlist *mrp, *nmrp;
7525 	struct nfslayout *lyp, *nlyp;
7526 
7527 	/* Get rid of the deviceid structures. */
7528 	nfsrv_freealldevids();
7529 	TAILQ_INIT(&nfsrv_devidhead);
7530 	nfsrv_devidcnt = 0;
7531 
7532 	/* Get rid of all layouts. */
7533 	nfsrv_freealllayouts();
7534 
7535 	/* Get rid of any nfsdontlist entries. */
7536 	LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7537 		free(mrp, M_NFSDSTATE);
7538 	LIST_INIT(&nfsrv_dontlisthead);
7539 	nfsrv_dontlistlen = 0;
7540 
7541 	/* Free layouts in the recall list. */
7542 	TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7543 		nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7544 	TAILQ_INIT(&nfsrv_recalllisthead);
7545 }
7546 
7547 /*
7548  * Free layouts that match the arguments.
7549  */
7550 static void
7551 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7552 {
7553 	struct nfslayouthash *lhyp;
7554 	struct nfslayout *lyp, *nlyp;
7555 	int i;
7556 
7557 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7558 		lhyp = &nfslayouthash[i];
7559 		NFSLOCKLAYOUT(lhyp);
7560 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7561 			if (clid->qval != lyp->lay_clientid.qval)
7562 				continue;
7563 			if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0)
7564 				continue;
7565 			if (laytype != lyp->lay_type)
7566 				continue;
7567 			if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7568 				lyp->lay_flags &= ~NFSLAY_READ;
7569 			if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7570 				lyp->lay_flags &= ~NFSLAY_RW;
7571 			if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7572 				nfsrv_freelayout(&lhyp->list, lyp);
7573 		}
7574 		NFSUNLOCKLAYOUT(lhyp);
7575 	}
7576 }
7577 
7578 /*
7579  * Free all layouts for the argument file.
7580  */
7581 void
7582 nfsrv_freefilelayouts(fhandle_t *fhp)
7583 {
7584 	struct nfslayouthash *lhyp;
7585 	struct nfslayout *lyp, *nlyp;
7586 
7587 	lhyp = NFSLAYOUTHASH(fhp);
7588 	NFSLOCKLAYOUT(lhyp);
7589 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7590 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7591 			nfsrv_freelayout(&lhyp->list, lyp);
7592 	}
7593 	NFSUNLOCKLAYOUT(lhyp);
7594 }
7595 
7596 /*
7597  * Free all layouts.
7598  */
7599 static void
7600 nfsrv_freealllayouts(void)
7601 {
7602 	struct nfslayouthash *lhyp;
7603 	struct nfslayout *lyp, *nlyp;
7604 	int i;
7605 
7606 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7607 		lhyp = &nfslayouthash[i];
7608 		NFSLOCKLAYOUT(lhyp);
7609 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7610 			nfsrv_freelayout(&lhyp->list, lyp);
7611 		NFSUNLOCKLAYOUT(lhyp);
7612 	}
7613 }
7614 
7615 /*
7616  * Look up the mount path for the DS server.
7617  */
7618 static int
7619 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7620     struct nfsdevice **dsp)
7621 {
7622 	struct nameidata nd;
7623 	struct nfsdevice *ds;
7624 	struct mount *mp;
7625 	int error, i;
7626 	char *dsdirpath;
7627 	size_t dsdirsize;
7628 
7629 	NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7630 	*dsp = NULL;
7631 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7632 	    dspathp, p);
7633 	error = namei(&nd);
7634 	NFSD_DEBUG(4, "lookup=%d\n", error);
7635 	if (error != 0)
7636 		return (error);
7637 	if (nd.ni_vp->v_type != VDIR) {
7638 		vput(nd.ni_vp);
7639 		NFSD_DEBUG(4, "dspath not dir\n");
7640 		return (ENOTDIR);
7641 	}
7642 	if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7643 		vput(nd.ni_vp);
7644 		NFSD_DEBUG(4, "dspath not an NFS mount\n");
7645 		return (ENXIO);
7646 	}
7647 
7648 	/*
7649 	 * Allocate a DS server structure with the NFS mounted directory
7650 	 * vnode reference counted, so that a non-forced dismount will
7651 	 * fail with EBUSY.
7652 	 * This structure is always linked into the list, even if an error
7653 	 * is being returned.  The caller will free the entire list upon
7654 	 * an error return.
7655 	 */
7656 	*dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7657 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7658 	ds->nfsdev_dvp = nd.ni_vp;
7659 	ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7660 	NFSVOPUNLOCK(nd.ni_vp);
7661 
7662 	dsdirsize = strlen(dspathp) + 16;
7663 	dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7664 	/* Now, create the DS directory structures. */
7665 	for (i = 0; i < nfsrv_dsdirsize; i++) {
7666 		snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7667 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7668 		    UIO_SYSSPACE, dsdirpath, p);
7669 		error = namei(&nd);
7670 		NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7671 		if (error != 0)
7672 			break;
7673 		if (nd.ni_vp->v_type != VDIR) {
7674 			vput(nd.ni_vp);
7675 			error = ENOTDIR;
7676 			NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7677 			break;
7678 		}
7679 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7680 			vput(nd.ni_vp);
7681 			error = ENXIO;
7682 			NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7683 			break;
7684 		}
7685 		ds->nfsdev_dsdir[i] = nd.ni_vp;
7686 		NFSVOPUNLOCK(nd.ni_vp);
7687 	}
7688 	free(dsdirpath, M_TEMP);
7689 
7690 	if (strlen(mdspathp) > 0) {
7691 		/*
7692 		 * This DS stores file for a specific MDS exported file
7693 		 * system.
7694 		 */
7695 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7696 		    UIO_SYSSPACE, mdspathp, p);
7697 		error = namei(&nd);
7698 		NFSD_DEBUG(4, "mds lookup=%d\n", error);
7699 		if (error != 0)
7700 			goto out;
7701 		if (nd.ni_vp->v_type != VDIR) {
7702 			vput(nd.ni_vp);
7703 			error = ENOTDIR;
7704 			NFSD_DEBUG(4, "mdspath not dir\n");
7705 			goto out;
7706 		}
7707 		mp = nd.ni_vp->v_mount;
7708 		if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7709 			vput(nd.ni_vp);
7710 			error = ENXIO;
7711 			NFSD_DEBUG(4, "mdspath not an exported fs\n");
7712 			goto out;
7713 		}
7714 		ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7715 		ds->nfsdev_mdsisset = 1;
7716 		vput(nd.ni_vp);
7717 	}
7718 
7719 out:
7720 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7721 	atomic_add_int(&nfsrv_devidcnt, 1);
7722 	return (error);
7723 }
7724 
7725 /*
7726  * Look up the mount path for the DS server and delete it.
7727  */
7728 int
7729 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7730 {
7731 	struct mount *mp;
7732 	struct nfsmount *nmp;
7733 	struct nfsdevice *ds;
7734 	int error;
7735 
7736 	NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7737 	/*
7738 	 * Search for the path in the mount list.  Avoid looking the path
7739 	 * up, since this mount point may be hung, with associated locked
7740 	 * vnodes, etc.
7741 	 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7742 	 * until this completes.
7743 	 * As noted in the man page, this should be done before any forced
7744 	 * dismount on the mount point, but at least the handshake on
7745 	 * NFSMNTP_CANCELRPCS should make it safe.
7746 	 */
7747 	error = 0;
7748 	ds = NULL;
7749 	nmp = NULL;
7750 	mtx_lock(&mountlist_mtx);
7751 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7752 		if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7753 		    strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7754 		    mp->mnt_data != NULL) {
7755 			nmp = VFSTONFS(mp);
7756 			NFSLOCKMNT(nmp);
7757 			if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7758 			     NFSMNTP_CANCELRPCS)) == 0) {
7759 				nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7760 				NFSUNLOCKMNT(nmp);
7761 			} else {
7762 				NFSUNLOCKMNT(nmp);
7763 				nmp = NULL;
7764 			}
7765 			break;
7766 		}
7767 	}
7768 	mtx_unlock(&mountlist_mtx);
7769 
7770 	if (nmp != NULL) {
7771 		ds = nfsrv_deldsnmp(op, nmp, p);
7772 		NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7773 		if (ds != NULL) {
7774 			nfsrv_killrpcs(nmp);
7775 			NFSD_DEBUG(4, "aft killrpcs\n");
7776 		} else
7777 			error = ENXIO;
7778 		NFSLOCKMNT(nmp);
7779 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7780 		wakeup(nmp);
7781 		NFSUNLOCKMNT(nmp);
7782 	} else
7783 		error = EINVAL;
7784 	return (error);
7785 }
7786 
7787 /*
7788  * Search for and remove a DS entry which matches the "nmp" argument.
7789  * The nfsdevice structure pointer is returned so that the caller can
7790  * free it via nfsrv_freeonedevid().
7791  * For the forced case, do not try to do LayoutRecalls, since the server
7792  * must be shut down now anyhow.
7793  */
7794 struct nfsdevice *
7795 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7796 {
7797 	struct nfsdevice *fndds;
7798 
7799 	NFSD_DEBUG(4, "deldsdvp\n");
7800 	NFSDDSLOCK();
7801 	if (op == PNFSDOP_FORCEDELDS)
7802 		fndds = nfsv4_findmirror(nmp);
7803 	else
7804 		fndds = nfsrv_findmirroredds(nmp);
7805 	if (fndds != NULL)
7806 		nfsrv_deleteds(fndds);
7807 	NFSDDSUNLOCK();
7808 	if (fndds != NULL) {
7809 		if (op != PNFSDOP_FORCEDELDS)
7810 			nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7811 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7812 	}
7813 	return (fndds);
7814 }
7815 
7816 /*
7817  * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7818  * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7819  * point.
7820  * Also, returns an error instead of the nfsdevice found.
7821  */
7822 int
7823 nfsrv_delds(char *devid, NFSPROC_T *p)
7824 {
7825 	struct nfsdevice *ds, *fndds;
7826 	struct nfsmount *nmp;
7827 	int fndmirror;
7828 
7829 	NFSD_DEBUG(4, "delds\n");
7830 	/*
7831 	 * Search the DS server list for a match with devid.
7832 	 * Remove the DS entry if found and there is a mirror.
7833 	 */
7834 	fndds = NULL;
7835 	nmp = NULL;
7836 	fndmirror = 0;
7837 	NFSDDSLOCK();
7838 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7839 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7840 		    ds->nfsdev_nmp != NULL) {
7841 			NFSD_DEBUG(4, "fnd main ds\n");
7842 			fndds = ds;
7843 			break;
7844 		}
7845 	}
7846 	if (fndds == NULL) {
7847 		NFSDDSUNLOCK();
7848 		return (ENXIO);
7849 	}
7850 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7851 		fndmirror = 1;
7852 	else if (fndds->nfsdev_mdsisset != 0) {
7853 		/* For the fsid is set case, search for a mirror. */
7854 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7855 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
7856 			    ds->nfsdev_mdsisset != 0 &&
7857 			    fsidcmp(&ds->nfsdev_mdsfsid,
7858 			    &fndds->nfsdev_mdsfsid) == 0) {
7859 				fndmirror = 1;
7860 				break;
7861 			}
7862 		}
7863 	}
7864 	if (fndmirror != 0) {
7865 		nmp = fndds->nfsdev_nmp;
7866 		NFSLOCKMNT(nmp);
7867 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7868 		     NFSMNTP_CANCELRPCS)) == 0) {
7869 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7870 			NFSUNLOCKMNT(nmp);
7871 			nfsrv_deleteds(fndds);
7872 		} else {
7873 			NFSUNLOCKMNT(nmp);
7874 			nmp = NULL;
7875 		}
7876 	}
7877 	NFSDDSUNLOCK();
7878 	if (nmp != NULL) {
7879 		nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7880 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7881 		nfsrv_killrpcs(nmp);
7882 		NFSLOCKMNT(nmp);
7883 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7884 		wakeup(nmp);
7885 		NFSUNLOCKMNT(nmp);
7886 		return (0);
7887 	}
7888 	return (ENXIO);
7889 }
7890 
7891 /*
7892  * Mark a DS as disabled by setting nfsdev_nmp = NULL.
7893  */
7894 static void
7895 nfsrv_deleteds(struct nfsdevice *fndds)
7896 {
7897 
7898 	NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
7899 	fndds->nfsdev_nmp = NULL;
7900 	if (fndds->nfsdev_mdsisset == 0)
7901 		nfsrv_faildscnt--;
7902 }
7903 
7904 /*
7905  * Fill in the addr structures for the File and Flex File layouts.
7906  */
7907 static void
7908 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
7909 {
7910 	uint32_t *tl;
7911 	char *netprot;
7912 	int addrlen;
7913 	static uint64_t new_devid = 0;
7914 
7915 	if (strchr(addr, ':') != NULL)
7916 		netprot = "tcp6";
7917 	else
7918 		netprot = "tcp";
7919 
7920 	/* Fill in the device id. */
7921 	NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
7922 	new_devid++;
7923 	NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
7924 	    sizeof(new_devid));
7925 
7926 	/*
7927 	 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
7928 	 * as defined in RFC5661) in XDR.
7929 	 */
7930 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7931 	    6 * NFSX_UNSIGNED;
7932 	NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
7933 	ds->nfsdev_fileaddrlen = addrlen;
7934 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7935 	ds->nfsdev_fileaddr = (char *)tl;
7936 	*tl++ = txdr_unsigned(1);		/* One stripe with index 0. */
7937 	*tl++ = 0;
7938 	*tl++ = txdr_unsigned(1);		/* One multipath list */
7939 	*tl++ = txdr_unsigned(1);		/* with one entry in it. */
7940 	/* The netaddr for this one entry. */
7941 	*tl++ = txdr_unsigned(strlen(netprot));
7942 	NFSBCOPY(netprot, tl, strlen(netprot));
7943 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7944 	*tl++ = txdr_unsigned(strlen(addr));
7945 	NFSBCOPY(addr, tl, strlen(addr));
7946 
7947 	/*
7948 	 * Fill in the flex file addr (actually the ff_device_addr4
7949 	 * as defined for Flexible File Layout) in XDR.
7950 	 */
7951 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7952 	    14 * NFSX_UNSIGNED;
7953 	ds->nfsdev_flexaddrlen = addrlen;
7954 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7955 	ds->nfsdev_flexaddr = (char *)tl;
7956 	*tl++ = txdr_unsigned(1);		/* One multipath entry. */
7957 	/* The netaddr for this one entry. */
7958 	*tl++ = txdr_unsigned(strlen(netprot));
7959 	NFSBCOPY(netprot, tl, strlen(netprot));
7960 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7961 	*tl++ = txdr_unsigned(strlen(addr));
7962 	NFSBCOPY(addr, tl, strlen(addr));
7963 	tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
7964 	*tl++ = txdr_unsigned(2);		/* Two NFS Versions. */
7965 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
7966 	*tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */
7967 	*tl++ = txdr_unsigned(NFS_SRVMAXIO);	/* DS max rsize. */
7968 	*tl++ = txdr_unsigned(NFS_SRVMAXIO);	/* DS max wsize. */
7969 	*tl++ = newnfs_true;			/* Tightly coupled. */
7970 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
7971 	*tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
7972 	*tl++ = txdr_unsigned(NFS_SRVMAXIO);	/* DS max rsize. */
7973 	*tl++ = txdr_unsigned(NFS_SRVMAXIO);	/* DS max wsize. */
7974 	*tl = newnfs_true;			/* Tightly coupled. */
7975 
7976 	ds->nfsdev_hostnamelen = strlen(dnshost);
7977 	ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
7978 	    M_WAITOK);
7979 	NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
7980 }
7981 
7982 /*
7983  * Create the device id list.
7984  * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
7985  * is misconfigured.
7986  */
7987 int
7988 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
7989 {
7990 	struct nfsdevice *ds;
7991 	char *addrp, *dnshostp, *dspathp, *mdspathp;
7992 	int error, i;
7993 
7994 	addrp = args->addr;
7995 	dnshostp = args->dnshost;
7996 	dspathp = args->dspath;
7997 	mdspathp = args->mdspath;
7998 	nfsrv_maxpnfsmirror = args->mirrorcnt;
7999 	if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
8000 	    mdspathp == NULL)
8001 		return (0);
8002 
8003 	/*
8004 	 * Loop around for each nul-terminated string in args->addr,
8005 	 * args->dnshost, args->dnspath and args->mdspath.
8006 	 */
8007 	while (addrp < (args->addr + args->addrlen) &&
8008 	    dnshostp < (args->dnshost + args->dnshostlen) &&
8009 	    dspathp < (args->dspath + args->dspathlen) &&
8010 	    mdspathp < (args->mdspath + args->mdspathlen)) {
8011 		error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
8012 		if (error != 0) {
8013 			/* Free all DS servers. */
8014 			nfsrv_freealldevids();
8015 			nfsrv_devidcnt = 0;
8016 			return (ENXIO);
8017 		}
8018 		nfsrv_allocdevid(ds, addrp, dnshostp);
8019 		addrp += (strlen(addrp) + 1);
8020 		dnshostp += (strlen(dnshostp) + 1);
8021 		dspathp += (strlen(dspathp) + 1);
8022 		mdspathp += (strlen(mdspathp) + 1);
8023 	}
8024 	if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
8025 		/* Free all DS servers. */
8026 		nfsrv_freealldevids();
8027 		nfsrv_devidcnt = 0;
8028 		nfsrv_maxpnfsmirror = 1;
8029 		return (ENXIO);
8030 	}
8031 	/* We can fail at most one less DS than the mirror level. */
8032 	nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
8033 
8034 	/*
8035 	 * Allocate the nfslayout hash table now, since this is a pNFS server.
8036 	 * Make it 1% of the high water mark and at least 100.
8037 	 */
8038 	if (nfslayouthash == NULL) {
8039 		nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
8040 		if (nfsrv_layouthashsize < 100)
8041 			nfsrv_layouthashsize = 100;
8042 		nfslayouthash = mallocarray(nfsrv_layouthashsize,
8043 		    sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
8044 		    M_ZERO);
8045 		for (i = 0; i < nfsrv_layouthashsize; i++) {
8046 			mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
8047 			TAILQ_INIT(&nfslayouthash[i].list);
8048 		}
8049 	}
8050 	return (0);
8051 }
8052 
8053 /*
8054  * Free all device ids.
8055  */
8056 static void
8057 nfsrv_freealldevids(void)
8058 {
8059 	struct nfsdevice *ds, *nds;
8060 
8061 	TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
8062 		nfsrv_freedevid(ds);
8063 }
8064 
8065 /*
8066  * Check to see if there is a Read/Write Layout plus either:
8067  * - A Write Delegation
8068  * or
8069  * - An Open with Write_access.
8070  * Return 1 if this is the case and 0 otherwise.
8071  * This function is used by nfsrv_proxyds() to decide if doing a Proxy
8072  * Getattr RPC to the Data Server (DS) is necessary.
8073  */
8074 #define	NFSCLIDVECSIZE	6
8075 int
8076 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p)
8077 {
8078 	fhandle_t fh, *tfhp;
8079 	struct nfsstate *stp;
8080 	struct nfslayout *lyp;
8081 	struct nfslayouthash *lhyp;
8082 	struct nfslockhashhead *hp;
8083 	struct nfslockfile *lfp;
8084 	nfsquad_t clid[NFSCLIDVECSIZE];
8085 	int clidcnt, ret;
8086 
8087 	ret = nfsvno_getfh(vp, &fh, p);
8088 	if (ret != 0)
8089 		return (0);
8090 
8091 	/* First check for a Read/Write Layout. */
8092 	clidcnt = 0;
8093 	lhyp = NFSLAYOUTHASH(&fh);
8094 	NFSLOCKLAYOUT(lhyp);
8095 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8096 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8097 		    ((lyp->lay_flags & NFSLAY_RW) != 0 ||
8098 		     ((lyp->lay_flags & NFSLAY_READ) != 0 &&
8099 		      nfsrv_pnfsatime != 0))) {
8100 			if (clidcnt < NFSCLIDVECSIZE)
8101 				clid[clidcnt].qval = lyp->lay_clientid.qval;
8102 			clidcnt++;
8103 		}
8104 	}
8105 	NFSUNLOCKLAYOUT(lhyp);
8106 	if (clidcnt == 0) {
8107 		/* None found, so return 0. */
8108 		return (0);
8109 	}
8110 
8111 	/* Get the nfslockfile for this fh. */
8112 	NFSLOCKSTATE();
8113 	hp = NFSLOCKHASH(&fh);
8114 	LIST_FOREACH(lfp, hp, lf_hash) {
8115 		tfhp = &lfp->lf_fh;
8116 		if (NFSVNO_CMPFH(&fh, tfhp))
8117 			break;
8118 	}
8119 	if (lfp == NULL) {
8120 		/* None found, so return 0. */
8121 		NFSUNLOCKSTATE();
8122 		return (0);
8123 	}
8124 
8125 	/* Now, look for a Write delegation for this clientid. */
8126 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
8127 		if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8128 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8129 			break;
8130 	}
8131 	if (stp != NULL) {
8132 		/* Found one, so return 1. */
8133 		NFSUNLOCKSTATE();
8134 		return (1);
8135 	}
8136 
8137 	/* No Write delegation, so look for an Open with Write_access. */
8138 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
8139 		KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
8140 		    ("nfsrv_checkdsattr: Non-open in Open list\n"));
8141 		if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
8142 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8143 			break;
8144 	}
8145 	NFSUNLOCKSTATE();
8146 	if (stp != NULL)
8147 		return (1);
8148 	return (0);
8149 }
8150 
8151 /*
8152  * Look for a matching clientid in the vector. Return 1 if one might match.
8153  */
8154 static int
8155 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
8156 {
8157 	int i;
8158 
8159 	/* If too many for the vector, return 1 since there might be a match. */
8160 	if (clidcnt > NFSCLIDVECSIZE)
8161 		return (1);
8162 
8163 	for (i = 0; i < clidcnt; i++)
8164 		if (clidvec[i].qval == clid.qval)
8165 			return (1);
8166 	return (0);
8167 }
8168 
8169 /*
8170  * Check the don't list for "vp" and see if issuing an rw layout is allowed.
8171  * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
8172  */
8173 static int
8174 nfsrv_dontlayout(fhandle_t *fhp)
8175 {
8176 	struct nfsdontlist *mrp;
8177 	int ret;
8178 
8179 	if (nfsrv_dontlistlen == 0)
8180 		return (0);
8181 	ret = 0;
8182 	NFSDDONTLISTLOCK();
8183 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8184 		if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
8185 		    (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
8186 			ret = 1;
8187 			break;
8188 		}
8189 	}
8190 	NFSDDONTLISTUNLOCK();
8191 	return (ret);
8192 }
8193 
8194 #define	PNFSDS_COPYSIZ	65536
8195 /*
8196  * Create a new file on a DS and copy the contents of an extant DS file to it.
8197  * This can be used for recovery of a DS file onto a recovered DS.
8198  * The steps are:
8199  * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
8200  * - Disable issuing of read/write layouts for the file via the nfsdontlist,
8201  *   so that they will be disabled after the MDS file's vnode is unlocked.
8202  * - Set up the nfsrv_recalllist so that recall of read/write layouts can
8203  *   be done.
8204  * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
8205  *   writes, LayoutCommits and LayoutReturns for the file when completing the
8206  *   LayoutReturn requested by the LayoutRecall callback.
8207  * - Issue a LayoutRecall callback for all read/write layouts and wait for
8208  *   them to be returned. (If the LayoutRecall callback replies
8209  *   NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
8210  * - Exclusively lock the MDS file's vnode.  This ensures that no proxied
8211  *   writes are in progress or can occur during the DS file copy.
8212  *   It also blocks Setattr operations.
8213  * - Create the file on the recovered mirror.
8214  * - Copy the file from the operational DS.
8215  * - Copy any ACL from the MDS file to the new DS file.
8216  * - Set the modify time of the new DS file to that of the MDS file.
8217  * - Update the extended attribute for the MDS file.
8218  * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
8219  * - The caller will unlock the MDS file's vnode allowing operations
8220  *   to continue normally, since it is now on the mirror again.
8221  */
8222 int
8223 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
8224     struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
8225     struct ucred *cred, NFSPROC_T *p)
8226 {
8227 	struct nfsdontlist *mrp, *nmrp;
8228 	struct nfslayouthash *lhyp;
8229 	struct nfslayout *lyp, *nlyp;
8230 	struct nfslayouthead thl;
8231 	struct mount *mp, *tvmp;
8232 	struct acl *aclp;
8233 	struct vattr va;
8234 	struct timespec mtime;
8235 	fhandle_t fh;
8236 	vnode_t tvp;
8237 	off_t rdpos, wrpos;
8238 	ssize_t aresid;
8239 	char *dat;
8240 	int didprintf, ret, retacl, xfer;
8241 
8242 	ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
8243 	ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
8244 	/*
8245 	 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
8246 	 * so that no more RW layouts will get issued.
8247 	 */
8248 	ret = nfsvno_getfh(vp, &fh, p);
8249 	if (ret != 0) {
8250 		NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8251 		return (ret);
8252 	}
8253 	nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8254 	nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8255 	NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8256 	NFSDDONTLISTLOCK();
8257 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8258 		if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8259 			break;
8260 	}
8261 	if (mrp == NULL) {
8262 		LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8263 		mrp = nmrp;
8264 		nmrp = NULL;
8265 		nfsrv_dontlistlen++;
8266 		NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8267 	} else {
8268 		NFSDDONTLISTUNLOCK();
8269 		free(nmrp, M_NFSDSTATE);
8270 		NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8271 		return (ENXIO);
8272 	}
8273 	NFSDDONTLISTUNLOCK();
8274 
8275 	/*
8276 	 * Search for all RW layouts for this file.  Move them to the
8277 	 * recall list, so they can be recalled and their return noted.
8278 	 */
8279 	lhyp = NFSLAYOUTHASH(&fh);
8280 	NFSDRECALLLOCK();
8281 	NFSLOCKLAYOUT(lhyp);
8282 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8283 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8284 		    (lyp->lay_flags & NFSLAY_RW) != 0) {
8285 			TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8286 			TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8287 			lyp->lay_trycnt = 0;
8288 		}
8289 	}
8290 	NFSUNLOCKLAYOUT(lhyp);
8291 	NFSDRECALLUNLOCK();
8292 
8293 	ret = 0;
8294 	mp = tvmp = NULL;
8295 	didprintf = 0;
8296 	TAILQ_INIT(&thl);
8297 	/* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8298 	NFSVOPUNLOCK(vp);
8299 	/* Now, do a recall for all layouts not yet recalled. */
8300 tryagain:
8301 	NFSDRECALLLOCK();
8302 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8303 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8304 		    (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8305 			lyp->lay_flags |= NFSLAY_RECALL;
8306 			/*
8307 			 * The layout stateid.seqid needs to be incremented
8308 			 * before doing a LAYOUT_RECALL callback.
8309 			 */
8310 			if (++lyp->lay_stateid.seqid == 0)
8311 				lyp->lay_stateid.seqid = 1;
8312 			NFSDRECALLUNLOCK();
8313 			nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8314 			    &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8315 			NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8316 			goto tryagain;
8317 		}
8318 	}
8319 
8320 	/* Now wait for them to be returned. */
8321 tryagain2:
8322 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8323 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8324 			if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8325 				TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8326 				    lay_list);
8327 				TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8328 				NFSD_DEBUG(4,
8329 				    "nfsrv_copymr: layout returned\n");
8330 			} else {
8331 				lyp->lay_trycnt++;
8332 				ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8333 				    PVFS | PCATCH, "nfsmrl", hz);
8334 				NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8335 				    ret);
8336 				if (ret == EINTR || ret == ERESTART)
8337 					break;
8338 				if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8339 					/*
8340 					 * Give up after 60sec and return
8341 					 * ENXIO, failing the copymr.
8342 					 * This layout will remain on the
8343 					 * recalllist.  It can only be cleared
8344 					 * by restarting the nfsd.
8345 					 * This seems the safe way to handle
8346 					 * it, since it cannot be safely copied
8347 					 * with an outstanding RW layout.
8348 					 */
8349 					if (lyp->lay_trycnt >= 60) {
8350 						ret = ENXIO;
8351 						break;
8352 					}
8353 					if (didprintf == 0) {
8354 						printf("nfsrv_copymr: layout "
8355 						    "not returned\n");
8356 						didprintf = 1;
8357 					}
8358 				}
8359 			}
8360 			goto tryagain2;
8361 		}
8362 	}
8363 	NFSDRECALLUNLOCK();
8364 	/* We can now get rid of the layouts that have been returned. */
8365 	TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8366 		nfsrv_freelayout(&thl, lyp);
8367 
8368 	/*
8369 	 * Do the vn_start_write() calls here, before the MDS vnode is
8370 	 * locked and the tvp is created (locked) in the NFS file system
8371 	 * that dvp is in.
8372 	 * For tvmp, this probably isn't necessary, since it will be an
8373 	 * NFS mount and they are not suspendable at this time.
8374 	 */
8375 	if (ret == 0)
8376 		ret = vn_start_write(vp, &mp, V_WAIT | PCATCH);
8377 	if (ret == 0) {
8378 		tvmp = dvp->v_mount;
8379 		ret = vn_start_write(NULL, &tvmp, V_WAIT | PCATCH);
8380 	}
8381 
8382 	/*
8383 	 * LK_EXCLUSIVE lock the MDS vnode, so that any
8384 	 * proxied writes through the MDS will be blocked until we have
8385 	 * completed the copy and update of the extended attributes.
8386 	 * This will also ensure that any attributes and ACL will not be
8387 	 * changed until the copy is complete.
8388 	 */
8389 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8390 	if (ret == 0 && VN_IS_DOOMED(vp)) {
8391 		NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8392 		ret = ESTALE;
8393 	}
8394 
8395 	/* Create the data file on the recovered DS. */
8396 	if (ret == 0)
8397 		ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8398 
8399 	/* Copy the DS file, if created successfully. */
8400 	if (ret == 0) {
8401 		/*
8402 		 * Get any NFSv4 ACL on the MDS file, so that it can be set
8403 		 * on the new DS file.
8404 		 */
8405 		aclp = acl_alloc(M_WAITOK | M_ZERO);
8406 		retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8407 		if (retacl != 0 && retacl != ENOATTR)
8408 			NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8409 		dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8410 		/* Malloc a block of 0s used to check for holes. */
8411 		if (nfsrv_zeropnfsdat == NULL)
8412 			nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8413 			    M_WAITOK | M_ZERO);
8414 		rdpos = wrpos = 0;
8415 		ret = VOP_GETATTR(fvp, &va, cred);
8416 		aresid = 0;
8417 		while (ret == 0 && aresid == 0) {
8418 			ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8419 			    rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8420 			    &aresid, p);
8421 			xfer = PNFSDS_COPYSIZ - aresid;
8422 			if (ret == 0 && xfer > 0) {
8423 				rdpos += xfer;
8424 				/*
8425 				 * Skip the write for holes, except for the
8426 				 * last block.
8427 				 */
8428 				if (xfer < PNFSDS_COPYSIZ || rdpos ==
8429 				    va.va_size || NFSBCMP(dat,
8430 				    nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8431 					ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8432 					    wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8433 					    cred, NULL, NULL, p);
8434 				if (ret == 0)
8435 					wrpos += xfer;
8436 			}
8437 		}
8438 
8439 		/* If there is an ACL and the copy succeeded, set the ACL. */
8440 		if (ret == 0 && retacl == 0) {
8441 			ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8442 			/*
8443 			 * Don't consider these as errors, since VOP_GETACL()
8444 			 * can return an ACL when they are not actually
8445 			 * supported.  For example, for UFS, VOP_GETACL()
8446 			 * will return a trivial ACL based on the uid/gid/mode
8447 			 * when there is no ACL on the file.
8448 			 * This case should be recognized as a trivial ACL
8449 			 * by UFS's VOP_SETACL() and succeed, but...
8450 			 */
8451 			if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8452 				ret = 0;
8453 		}
8454 
8455 		if (ret == 0)
8456 			ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8457 
8458 		/* Set the DS data file's modify time that of the MDS file. */
8459 		if (ret == 0)
8460 			ret = VOP_GETATTR(vp, &va, cred);
8461 		if (ret == 0) {
8462 			mtime = va.va_mtime;
8463 			VATTR_NULL(&va);
8464 			va.va_mtime = mtime;
8465 			ret = VOP_SETATTR(tvp, &va, cred);
8466 		}
8467 
8468 		vput(tvp);
8469 		acl_free(aclp);
8470 		free(dat, M_TEMP);
8471 	}
8472 	if (tvmp != NULL)
8473 		vn_finished_write(tvmp);
8474 
8475 	/* Update the extended attributes for the newly created DS file. */
8476 	if (ret == 0)
8477 		ret = vn_extattr_set(vp, IO_NODELOCKED,
8478 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8479 		    sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8480 	if (mp != NULL)
8481 		vn_finished_write(mp);
8482 
8483 	/* Get rid of the dontlist entry, so that Layouts can be issued. */
8484 	NFSDDONTLISTLOCK();
8485 	LIST_REMOVE(mrp, nfsmr_list);
8486 	NFSDDONTLISTUNLOCK();
8487 	free(mrp, M_NFSDSTATE);
8488 	return (ret);
8489 }
8490 
8491 /*
8492  * Create a data storage file on the recovered DS.
8493  */
8494 static int
8495 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8496     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8497     vnode_t *tvpp)
8498 {
8499 	struct vattr va, nva;
8500 	int error;
8501 
8502 	/* Make data file name based on FH. */
8503 	error = VOP_GETATTR(vp, &va, cred);
8504 	if (error == 0) {
8505 		/* Set the attributes for "vp" to Setattr the DS vp. */
8506 		VATTR_NULL(&nva);
8507 		nva.va_uid = va.va_uid;
8508 		nva.va_gid = va.va_gid;
8509 		nva.va_mode = va.va_mode;
8510 		nva.va_size = 0;
8511 		VATTR_NULL(&va);
8512 		va.va_type = VREG;
8513 		va.va_mode = nva.va_mode;
8514 		NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8515 		error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8516 		    pf->dsf_filename, cred, p, tvpp);
8517 	}
8518 	return (error);
8519 }
8520 
8521 /*
8522  * Look up the MDS file shared locked, and then get the extended attribute
8523  * to find the extant DS file to be copied to the new mirror.
8524  * If successful, *vpp is set to the MDS file's vp and *nvpp is
8525  * set to a DS data file for the MDS file, both exclusively locked.
8526  * The "buf" argument has the pnfsdsfile structure from the MDS file
8527  * in it and buflen is set to its length.
8528  */
8529 int
8530 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8531     int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8532     struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8533     struct nfsdevice **fdsp)
8534 {
8535 	struct nameidata nd;
8536 	struct vnode *vp, *curvp;
8537 	struct pnfsdsfile *pf;
8538 	struct nfsmount *nmp, *curnmp;
8539 	int dsdir, error, mirrorcnt, ippos;
8540 
8541 	vp = NULL;
8542 	curvp = NULL;
8543 	curnmp = NULL;
8544 	*dsp = NULL;
8545 	*fdsp = NULL;
8546 	if (dspathp == NULL && curdspathp != NULL)
8547 		return (EPERM);
8548 
8549 	/*
8550 	 * Look up the MDS file shared locked.  The lock will be upgraded
8551 	 * to an exclusive lock after any rw layouts have been returned.
8552 	 */
8553 	NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8554 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8555 	    mdspathp, p);
8556 	error = namei(&nd);
8557 	NFSD_DEBUG(4, "lookup=%d\n", error);
8558 	if (error != 0)
8559 		return (error);
8560 	if (nd.ni_vp->v_type != VREG) {
8561 		vput(nd.ni_vp);
8562 		NFSD_DEBUG(4, "mdspath not reg\n");
8563 		return (EISDIR);
8564 	}
8565 	vp = nd.ni_vp;
8566 
8567 	if (curdspathp != NULL) {
8568 		/*
8569 		 * Look up the current DS path and find the nfsdev structure for
8570 		 * it.
8571 		 */
8572 		NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8573 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8574 		    UIO_SYSSPACE, curdspathp, p);
8575 		error = namei(&nd);
8576 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8577 		if (error != 0) {
8578 			vput(vp);
8579 			return (error);
8580 		}
8581 		if (nd.ni_vp->v_type != VDIR) {
8582 			vput(nd.ni_vp);
8583 			vput(vp);
8584 			NFSD_DEBUG(4, "curdspath not dir\n");
8585 			return (ENOTDIR);
8586 		}
8587 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8588 			vput(nd.ni_vp);
8589 			vput(vp);
8590 			NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8591 			return (ENXIO);
8592 		}
8593 		curnmp = VFSTONFS(nd.ni_vp->v_mount);
8594 
8595 		/* Search the nfsdev list for a match. */
8596 		NFSDDSLOCK();
8597 		*fdsp = nfsv4_findmirror(curnmp);
8598 		NFSDDSUNLOCK();
8599 		if (*fdsp == NULL)
8600 			curnmp = NULL;
8601 		if (curnmp == NULL) {
8602 			vput(nd.ni_vp);
8603 			vput(vp);
8604 			NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8605 			return (ENXIO);
8606 		}
8607 		curvp = nd.ni_vp;
8608 	}
8609 
8610 	if (dspathp != NULL) {
8611 		/* Look up the nfsdev path and find the nfsdev structure. */
8612 		NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8613 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8614 		    UIO_SYSSPACE, dspathp, p);
8615 		error = namei(&nd);
8616 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8617 		if (error != 0) {
8618 			vput(vp);
8619 			if (curvp != NULL)
8620 				vput(curvp);
8621 			return (error);
8622 		}
8623 		if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8624 			vput(nd.ni_vp);
8625 			vput(vp);
8626 			if (curvp != NULL)
8627 				vput(curvp);
8628 			NFSD_DEBUG(4, "dspath not dir\n");
8629 			if (nd.ni_vp == curvp)
8630 				return (EPERM);
8631 			return (ENOTDIR);
8632 		}
8633 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8634 			vput(nd.ni_vp);
8635 			vput(vp);
8636 			if (curvp != NULL)
8637 				vput(curvp);
8638 			NFSD_DEBUG(4, "dspath not an NFS mount\n");
8639 			return (ENXIO);
8640 		}
8641 		nmp = VFSTONFS(nd.ni_vp->v_mount);
8642 
8643 		/*
8644 		 * Search the nfsdevice list for a match.  If curnmp == NULL,
8645 		 * this is a recovery and there must be a mirror.
8646 		 */
8647 		NFSDDSLOCK();
8648 		if (curnmp == NULL)
8649 			*dsp = nfsrv_findmirroredds(nmp);
8650 		else
8651 			*dsp = nfsv4_findmirror(nmp);
8652 		NFSDDSUNLOCK();
8653 		if (*dsp == NULL) {
8654 			vput(nd.ni_vp);
8655 			vput(vp);
8656 			if (curvp != NULL)
8657 				vput(curvp);
8658 			NFSD_DEBUG(4, "mdscopymr: no ds\n");
8659 			return (ENXIO);
8660 		}
8661 	} else {
8662 		nd.ni_vp = NULL;
8663 		nmp = NULL;
8664 	}
8665 
8666 	/*
8667 	 * Get a vp for an available DS data file using the extended
8668 	 * attribute on the MDS file.
8669 	 * If there is a valid entry for the new DS in the extended attribute
8670 	 * on the MDS file (as checked via the nmp argument),
8671 	 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8672 	 */
8673 	error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p,
8674 	    NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8675 	if (curvp != NULL)
8676 		vput(curvp);
8677 	if (nd.ni_vp == NULL) {
8678 		if (error == 0 && nmp != NULL) {
8679 			/* Search the nfsdev list for a match. */
8680 			NFSDDSLOCK();
8681 			*dsp = nfsrv_findmirroredds(nmp);
8682 			NFSDDSUNLOCK();
8683 		}
8684 		if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8685 			if (nvpp != NULL && *nvpp != NULL) {
8686 				vput(*nvpp);
8687 				*nvpp = NULL;
8688 			}
8689 			error = ENXIO;
8690 		}
8691 	} else
8692 		vput(nd.ni_vp);
8693 
8694 	/*
8695 	 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8696 	 * and is only allowed if there is a 0.0.0.0 IP address entry.
8697 	 * When curdspathp != NULL, the ippos will be set to that entry.
8698 	 */
8699 	if (error == 0 && dspathp != NULL && ippos == -1) {
8700 		if (nvpp != NULL && *nvpp != NULL) {
8701 			vput(*nvpp);
8702 			*nvpp = NULL;
8703 		}
8704 		error = ENXIO;
8705 	}
8706 	if (error == 0) {
8707 		*vpp = vp;
8708 
8709 		pf = (struct pnfsdsfile *)buf;
8710 		if (ippos == -1) {
8711 			/* If no zeroip pnfsdsfile, add one. */
8712 			ippos = *buflenp / sizeof(*pf);
8713 			*buflenp += sizeof(*pf);
8714 			pf += ippos;
8715 			pf->dsf_dir = dsdir;
8716 			strlcpy(pf->dsf_filename, fname,
8717 			    sizeof(pf->dsf_filename));
8718 		} else
8719 			pf += ippos;
8720 		*pfp = pf;
8721 	} else
8722 		vput(vp);
8723 	return (error);
8724 }
8725 
8726 /*
8727  * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8728  * Return one if found, NULL otherwise.
8729  */
8730 static struct nfsdevice *
8731 nfsrv_findmirroredds(struct nfsmount *nmp)
8732 {
8733 	struct nfsdevice *ds, *fndds;
8734 	int fndmirror;
8735 
8736 	mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8737 	/*
8738 	 * Search the DS server list for a match with nmp.
8739 	 * Remove the DS entry if found and there is a mirror.
8740 	 */
8741 	fndds = NULL;
8742 	fndmirror = 0;
8743 	if (nfsrv_devidcnt == 0)
8744 		return (fndds);
8745 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8746 		if (ds->nfsdev_nmp == nmp) {
8747 			NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8748 			fndds = ds;
8749 			break;
8750 		}
8751 	}
8752 	if (fndds == NULL)
8753 		return (fndds);
8754 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8755 		fndmirror = 1;
8756 	else if (fndds->nfsdev_mdsisset != 0) {
8757 		/* For the fsid is set case, search for a mirror. */
8758 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8759 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
8760 			    ds->nfsdev_mdsisset != 0 &&
8761 			    fsidcmp(&ds->nfsdev_mdsfsid,
8762 			    &fndds->nfsdev_mdsfsid) == 0) {
8763 				fndmirror = 1;
8764 				break;
8765 			}
8766 		}
8767 	}
8768 	if (fndmirror == 0) {
8769 		NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");
8770 		return (NULL);
8771 	}
8772 	return (fndds);
8773 }
8774