nfs_nfsdstate.c revision 269398
1/*-
2 * Copyright (c) 2009 Rick Macklem, University of Guelph
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 */
27
28#include <sys/cdefs.h>
29__FBSDID("$FreeBSD: stable/10/sys/fs/nfsserver/nfs_nfsdstate.c 269398 2014-08-01 21:10:41Z rmacklem $");
30
31#ifndef APPLEKEXT
32#include <fs/nfs/nfsport.h>
33
34struct nfsrv_stablefirst nfsrv_stablefirst;
35int nfsrv_issuedelegs = 0;
36int nfsrv_dolocallocks = 0;
37struct nfsv4lock nfsv4rootfs_lock;
38
39extern int newnfs_numnfsd;
40extern struct nfsstats newnfsstats;
41extern int nfsrv_lease;
42extern struct timeval nfsboottime;
43extern u_int32_t newnfs_true, newnfs_false;
44NFSV4ROOTLOCKMUTEX;
45NFSSTATESPINLOCK;
46
47/*
48 * Hash lists for nfs V4.
49 * (Some would put them in the .h file, but I don't like declaring storage
50 *  in a .h)
51 */
52struct nfsclienthashhead nfsclienthash[NFSCLIENTHASHSIZE];
53struct nfslockhashhead nfslockhash[NFSLOCKHASHSIZE];
54struct nfssessionhash nfssessionhash[NFSSESSIONHASHSIZE];
55#endif	/* !APPLEKEXT */
56
57static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
58static time_t nfsrvboottime;
59static int nfsrv_writedelegifpos = 1;
60static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
61static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
62static int nfsrv_nogsscallback = 0;
63
64/* local functions */
65static void nfsrv_dumpaclient(struct nfsclient *clp,
66    struct nfsd_dumpclients *dumpp);
67static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
68    NFSPROC_T *p);
69static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
70    NFSPROC_T *p);
71static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
72    NFSPROC_T *p);
73static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
74    int cansleep, NFSPROC_T *p);
75static void nfsrv_freenfslock(struct nfslock *lop);
76static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
77static void nfsrv_freedeleg(struct nfsstate *);
78static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
79    u_int32_t flags, struct nfsstate **stpp);
80static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
81    struct nfsstate **stpp);
82static int nfsrv_getlockfh(vnode_t vp, u_short flags,
83    struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
84static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
85    struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
86static void nfsrv_insertlock(struct nfslock *new_lop,
87    struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
88static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
89    struct nfslock **other_lopp, struct nfslockfile *lfp);
90static int nfsrv_getipnumber(u_char *cp);
91static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
92    nfsv4stateid_t *stateidp, int specialid);
93static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
94    u_int32_t flags);
95static int nfsrv_docallback(struct nfsclient *clp, int procnum,
96    nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
97    struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
98static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
99    uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
100static u_int32_t nfsrv_nextclientindex(void);
101static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
102static void nfsrv_markstable(struct nfsclient *clp);
103static int nfsrv_checkstable(struct nfsclient *clp);
104static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
105    vnode *vp, NFSPROC_T *p);
106static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
107    NFSPROC_T *p, vnode_t vp);
108static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
109    struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
110static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
111    struct nfsclient *clp);
112static time_t nfsrv_leaseexpiry(void);
113static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
114static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
115    struct nfsstate *stp, struct nfsrvcache *op);
116static int nfsrv_nootherstate(struct nfsstate *stp);
117static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
118    uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
119static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
120    uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
121static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
122    int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
123    NFSPROC_T *p);
124static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
125    NFSPROC_T *p);
126static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
127    uint64_t first, uint64_t end);
128static void nfsrv_locklf(struct nfslockfile *lfp);
129static void nfsrv_unlocklf(struct nfslockfile *lfp);
130static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
131static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
132static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
133    int dont_replycache, struct nfsdsession **sepp);
134static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
135
136/*
137 * Scan the client list for a match and either return the current one,
138 * create a new entry or return an error.
139 * If returning a non-error, the clp structure must either be linked into
140 * the client list or free'd.
141 */
142APPLESTATIC int
143nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
144    nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
145{
146	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
147	int i, error = 0;
148	struct nfsstate *stp, *tstp;
149	struct sockaddr_in *sad, *rad;
150	int zapit = 0, gotit, hasstate = 0, igotlock;
151	static u_int64_t confirm_index = 0;
152
153	/*
154	 * Check for state resource limit exceeded.
155	 */
156	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
157		error = NFSERR_RESOURCE;
158		goto out;
159	}
160
161	if (nfsrv_issuedelegs == 0 ||
162	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
163		/*
164		 * Don't do callbacks when delegations are disabled or
165		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
166		 * If establishing a callback connection is attempted
167		 * when a firewall is blocking the callback path, the
168		 * server may wait too long for the connect attempt to
169		 * succeed during the Open. Some clients, such as Linux,
170		 * may timeout and give up on the Open before the server
171		 * replies. Also, since AUTH_GSS callbacks are not
172		 * yet interoperability tested, they might cause the
173		 * server to crap out, if they get past the Init call to
174		 * the client.
175		 */
176		new_clp->lc_program = 0;
177
178	/* Lock out other nfsd threads */
179	NFSLOCKV4ROOTMUTEX();
180	nfsv4_relref(&nfsv4rootfs_lock);
181	do {
182		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
183		    NFSV4ROOTLOCKMUTEXPTR, NULL);
184	} while (!igotlock);
185	NFSUNLOCKV4ROOTMUTEX();
186
187	/*
188	 * Search for a match in the client list.
189	 */
190	gotit = i = 0;
191	while (i < NFSCLIENTHASHSIZE && !gotit) {
192	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
193		if (new_clp->lc_idlen == clp->lc_idlen &&
194		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
195			gotit = 1;
196			break;
197		}
198	    }
199	    i++;
200	}
201	if (!gotit ||
202	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
203		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
204			/*
205			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
206			 * client is trying to update a confirmed clientid.
207			 */
208			NFSLOCKV4ROOTMUTEX();
209			nfsv4_unlock(&nfsv4rootfs_lock, 1);
210			NFSUNLOCKV4ROOTMUTEX();
211			confirmp->lval[1] = 0;
212			error = NFSERR_NOENT;
213			goto out;
214		}
215		/*
216		 * Get rid of the old one.
217		 */
218		if (i != NFSCLIENTHASHSIZE) {
219			LIST_REMOVE(clp, lc_hash);
220			nfsrv_cleanclient(clp, p);
221			nfsrv_freedeleglist(&clp->lc_deleg);
222			nfsrv_freedeleglist(&clp->lc_olddeleg);
223			zapit = 1;
224		}
225		/*
226		 * Add it after assigning a client id to it.
227		 */
228		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
229		if ((nd->nd_flag & ND_NFSV41) != 0)
230			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
231			    ++confirm_index;
232		else
233			confirmp->qval = new_clp->lc_confirm.qval =
234			    ++confirm_index;
235		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
236		    (u_int32_t)nfsrvboottime;
237		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
238		    nfsrv_nextclientindex();
239		new_clp->lc_stateindex = 0;
240		new_clp->lc_statemaxindex = 0;
241		new_clp->lc_cbref = 0;
242		new_clp->lc_expiry = nfsrv_leaseexpiry();
243		LIST_INIT(&new_clp->lc_open);
244		LIST_INIT(&new_clp->lc_deleg);
245		LIST_INIT(&new_clp->lc_olddeleg);
246		LIST_INIT(&new_clp->lc_session);
247		for (i = 0; i < NFSSTATEHASHSIZE; i++)
248			LIST_INIT(&new_clp->lc_stateid[i]);
249		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
250		    lc_hash);
251		newnfsstats.srvclients++;
252		nfsrv_openpluslock++;
253		nfsrv_clients++;
254		NFSLOCKV4ROOTMUTEX();
255		nfsv4_unlock(&nfsv4rootfs_lock, 1);
256		NFSUNLOCKV4ROOTMUTEX();
257		if (zapit)
258			nfsrv_zapclient(clp, p);
259		*new_clpp = NULL;
260		goto out;
261	}
262
263	/*
264	 * Now, handle the cases where the id is already issued.
265	 */
266	if (nfsrv_notsamecredname(nd, clp)) {
267	    /*
268	     * Check to see if there is expired state that should go away.
269	     */
270	    if (clp->lc_expiry < NFSD_MONOSEC &&
271	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
272		nfsrv_cleanclient(clp, p);
273		nfsrv_freedeleglist(&clp->lc_deleg);
274	    }
275
276	    /*
277	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
278	     * RFC3530 Sec. 8.1.2 last para.
279	     */
280	    if (!LIST_EMPTY(&clp->lc_deleg)) {
281		hasstate = 1;
282	    } else if (LIST_EMPTY(&clp->lc_open)) {
283		hasstate = 0;
284	    } else {
285		hasstate = 0;
286		/* Look for an Open on the OpenOwner */
287		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
288		    if (!LIST_EMPTY(&stp->ls_open)) {
289			hasstate = 1;
290			break;
291		    }
292		}
293	    }
294	    if (hasstate) {
295		/*
296		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
297		 * filling out the correct ipaddr and portnum.
298		 */
299		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
300		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
301		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
302		sad->sin_port = rad->sin_port;
303		NFSLOCKV4ROOTMUTEX();
304		nfsv4_unlock(&nfsv4rootfs_lock, 1);
305		NFSUNLOCKV4ROOTMUTEX();
306		error = NFSERR_CLIDINUSE;
307		goto out;
308	    }
309	}
310
311	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
312		/*
313		 * If the verifier has changed, the client has rebooted
314		 * and a new client id is issued. The old state info
315		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
316		 */
317		LIST_REMOVE(clp, lc_hash);
318		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
319		if ((nd->nd_flag & ND_NFSV41) != 0)
320			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
321			    ++confirm_index;
322		else
323			confirmp->qval = new_clp->lc_confirm.qval =
324			    ++confirm_index;
325		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
326		    nfsrvboottime;
327		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
328		    nfsrv_nextclientindex();
329		new_clp->lc_stateindex = 0;
330		new_clp->lc_statemaxindex = 0;
331		new_clp->lc_cbref = 0;
332		new_clp->lc_expiry = nfsrv_leaseexpiry();
333
334		/*
335		 * Save the state until confirmed.
336		 */
337		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
338		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
339			tstp->ls_clp = new_clp;
340		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
341		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
342			tstp->ls_clp = new_clp;
343		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
344		    ls_list);
345		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
346			tstp->ls_clp = new_clp;
347		for (i = 0; i < NFSSTATEHASHSIZE; i++) {
348			LIST_NEWHEAD(&new_clp->lc_stateid[i],
349			    &clp->lc_stateid[i], ls_hash);
350			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
351				tstp->ls_clp = new_clp;
352		}
353		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
354		    lc_hash);
355		newnfsstats.srvclients++;
356		nfsrv_openpluslock++;
357		nfsrv_clients++;
358		NFSLOCKV4ROOTMUTEX();
359		nfsv4_unlock(&nfsv4rootfs_lock, 1);
360		NFSUNLOCKV4ROOTMUTEX();
361
362		/*
363		 * Must wait until any outstanding callback on the old clp
364		 * completes.
365		 */
366		NFSLOCKSTATE();
367		while (clp->lc_cbref) {
368			clp->lc_flags |= LCL_WAKEUPWANTED;
369			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
370			    "nfsd clp", 10 * hz);
371		}
372		NFSUNLOCKSTATE();
373		nfsrv_zapclient(clp, p);
374		*new_clpp = NULL;
375		goto out;
376	}
377
378	/* For NFSv4.1, mark that we found a confirmed clientid. */
379	if ((nd->nd_flag & ND_NFSV41) != 0)
380		confirmp->lval[1] = 1;
381	else {
382		/*
383		 * id and verifier match, so update the net address info
384		 * and get rid of any existing callback authentication
385		 * handle, so a new one will be acquired.
386		 */
387		LIST_REMOVE(clp, lc_hash);
388		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
389		new_clp->lc_expiry = nfsrv_leaseexpiry();
390		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
391		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
392		    clp->lc_clientid.lval[0];
393		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
394		    clp->lc_clientid.lval[1];
395		new_clp->lc_delegtime = clp->lc_delegtime;
396		new_clp->lc_stateindex = clp->lc_stateindex;
397		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
398		new_clp->lc_cbref = 0;
399		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
400		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
401			tstp->ls_clp = new_clp;
402		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
403		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
404			tstp->ls_clp = new_clp;
405		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
406		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
407			tstp->ls_clp = new_clp;
408		for (i = 0; i < NFSSTATEHASHSIZE; i++) {
409			LIST_NEWHEAD(&new_clp->lc_stateid[i],
410			    &clp->lc_stateid[i], ls_hash);
411			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
412				tstp->ls_clp = new_clp;
413		}
414		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
415		    lc_hash);
416		newnfsstats.srvclients++;
417		nfsrv_openpluslock++;
418		nfsrv_clients++;
419	}
420	NFSLOCKV4ROOTMUTEX();
421	nfsv4_unlock(&nfsv4rootfs_lock, 1);
422	NFSUNLOCKV4ROOTMUTEX();
423
424	if ((nd->nd_flag & ND_NFSV41) == 0) {
425		/*
426		 * Must wait until any outstanding callback on the old clp
427		 * completes.
428		 */
429		NFSLOCKSTATE();
430		while (clp->lc_cbref) {
431			clp->lc_flags |= LCL_WAKEUPWANTED;
432			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
433			    "nfsdclp", 10 * hz);
434		}
435		NFSUNLOCKSTATE();
436		nfsrv_zapclient(clp, p);
437		*new_clpp = NULL;
438	}
439
440out:
441	NFSEXITCODE2(error, nd);
442	return (error);
443}
444
445/*
446 * Check to see if the client id exists and optionally confirm it.
447 */
448APPLESTATIC int
449nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
450    struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
451    struct nfsrv_descript *nd, NFSPROC_T *p)
452{
453	struct nfsclient *clp;
454	struct nfsstate *stp;
455	int i;
456	struct nfsclienthashhead *hp;
457	int error = 0, igotlock, doneok;
458	struct nfssessionhash *shp;
459	struct nfsdsession *sep;
460	uint64_t sessid[2];
461	static uint64_t next_sess = 0;
462
463	if (clpp)
464		*clpp = NULL;
465	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
466	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
467		error = NFSERR_STALECLIENTID;
468		goto out;
469	}
470
471	/*
472	 * If called with opflags == CLOPS_RENEW, the State Lock is
473	 * already held. Otherwise, we need to get either that or,
474	 * for the case of Confirm, lock out the nfsd threads.
475	 */
476	if (opflags & CLOPS_CONFIRM) {
477		NFSLOCKV4ROOTMUTEX();
478		nfsv4_relref(&nfsv4rootfs_lock);
479		do {
480			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
481			    NFSV4ROOTLOCKMUTEXPTR, NULL);
482		} while (!igotlock);
483		/*
484		 * Create a new sessionid here, since we need to do it where
485		 * there is a mutex held to serialize update of next_sess.
486		 */
487		if ((nd->nd_flag & ND_NFSV41) != 0) {
488			sessid[0] = ++next_sess;
489			sessid[1] = clientid.qval;
490		}
491		NFSUNLOCKV4ROOTMUTEX();
492	} else if (opflags != CLOPS_RENEW) {
493		NFSLOCKSTATE();
494	}
495
496	/* For NFSv4.1, the clp is acquired from the associated session. */
497	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
498	    opflags == CLOPS_RENEW) {
499		clp = NULL;
500		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
501			shp = NFSSESSIONHASH(nd->nd_sessionid);
502			NFSLOCKSESSION(shp);
503			sep = nfsrv_findsession(nd->nd_sessionid);
504			if (sep != NULL)
505				clp = sep->sess_clp;
506			NFSUNLOCKSESSION(shp);
507		}
508	} else {
509		hp = NFSCLIENTHASH(clientid);
510		LIST_FOREACH(clp, hp, lc_hash) {
511			if (clp->lc_clientid.lval[1] == clientid.lval[1])
512				break;
513		}
514	}
515	if (clp == NULL) {
516		if (opflags & CLOPS_CONFIRM)
517			error = NFSERR_STALECLIENTID;
518		else
519			error = NFSERR_EXPIRED;
520	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
521		/*
522		 * If marked admin revoked, just return the error.
523		 */
524		error = NFSERR_ADMINREVOKED;
525	}
526	if (error) {
527		if (opflags & CLOPS_CONFIRM) {
528			NFSLOCKV4ROOTMUTEX();
529			nfsv4_unlock(&nfsv4rootfs_lock, 1);
530			NFSUNLOCKV4ROOTMUTEX();
531		} else if (opflags != CLOPS_RENEW) {
532			NFSUNLOCKSTATE();
533		}
534		goto out;
535	}
536
537	/*
538	 * Perform any operations specified by the opflags.
539	 */
540	if (opflags & CLOPS_CONFIRM) {
541		if (((nd->nd_flag & ND_NFSV41) != 0 &&
542		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
543		    ((nd->nd_flag & ND_NFSV41) == 0 &&
544		     clp->lc_confirm.qval != confirm.qval))
545			error = NFSERR_STALECLIENTID;
546		else if (nfsrv_notsamecredname(nd, clp))
547			error = NFSERR_CLIDINUSE;
548
549		if (!error) {
550		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
551			LCL_NEEDSCONFIRM) {
552			/*
553			 * Hang onto the delegations (as old delegations)
554			 * for an Open with CLAIM_DELEGATE_PREV unless in
555			 * grace, but get rid of the rest of the state.
556			 */
557			nfsrv_cleanclient(clp, p);
558			nfsrv_freedeleglist(&clp->lc_olddeleg);
559			if (nfsrv_checkgrace(nd, clp, 0)) {
560			    /* In grace, so just delete delegations */
561			    nfsrv_freedeleglist(&clp->lc_deleg);
562			} else {
563			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
564				stp->ls_flags |= NFSLCK_OLDDELEG;
565			    clp->lc_delegtime = NFSD_MONOSEC +
566				nfsrv_lease + NFSRV_LEASEDELTA;
567			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
568				ls_list);
569			}
570			if ((nd->nd_flag & ND_NFSV41) != 0)
571			    clp->lc_program = cbprogram;
572		    }
573		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
574		    if (clp->lc_program)
575			clp->lc_flags |= LCL_NEEDSCBNULL;
576		    /* For NFSv4.1, link the session onto the client. */
577		    if (nsep != NULL) {
578			/* Hold a reference on the xprt for a backchannel. */
579			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
580			    != 0 && clp->lc_req.nr_client == NULL) {
581			    clp->lc_req.nr_client = (struct __rpc_client *)
582				clnt_bck_create(nd->nd_xprt->xp_socket,
583				cbprogram, NFSV4_CBVERS);
584			    if (clp->lc_req.nr_client != NULL) {
585				SVC_ACQUIRE(nd->nd_xprt);
586				nd->nd_xprt->xp_p2 =
587				    clp->lc_req.nr_client->cl_private;
588				/* Disable idle timeout. */
589				nd->nd_xprt->xp_idletimeout = 0;
590				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
591			    } else
592				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
593			}
594			NFSBCOPY(sessid, nsep->sess_sessionid,
595			    NFSX_V4SESSIONID);
596			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
597			    NFSX_V4SESSIONID);
598			shp = NFSSESSIONHASH(nsep->sess_sessionid);
599			NFSLOCKSESSION(shp);
600			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
601			NFSLOCKSTATE();
602			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
603			nsep->sess_clp = clp;
604			NFSUNLOCKSTATE();
605			NFSUNLOCKSESSION(shp);
606		    }
607		}
608	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
609		error = NFSERR_EXPIRED;
610	}
611
612	/*
613	 * If called by the Renew Op, we must check the principal.
614	 */
615	if (!error && (opflags & CLOPS_RENEWOP)) {
616	    if (nfsrv_notsamecredname(nd, clp)) {
617		doneok = 0;
618		for (i = 0; i < NFSSTATEHASHSIZE && doneok == 0; i++) {
619		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
620			if ((stp->ls_flags & NFSLCK_OPEN) &&
621			    stp->ls_uid == nd->nd_cred->cr_uid) {
622				doneok = 1;
623				break;
624			}
625		    }
626		}
627		if (!doneok)
628			error = NFSERR_ACCES;
629	    }
630	    if (!error && (clp->lc_flags & LCL_CBDOWN))
631		error = NFSERR_CBPATHDOWN;
632	}
633	if ((!error || error == NFSERR_CBPATHDOWN) &&
634	     (opflags & CLOPS_RENEW)) {
635		clp->lc_expiry = nfsrv_leaseexpiry();
636	}
637	if (opflags & CLOPS_CONFIRM) {
638		NFSLOCKV4ROOTMUTEX();
639		nfsv4_unlock(&nfsv4rootfs_lock, 1);
640		NFSUNLOCKV4ROOTMUTEX();
641	} else if (opflags != CLOPS_RENEW) {
642		NFSUNLOCKSTATE();
643	}
644	if (clpp)
645		*clpp = clp;
646
647out:
648	NFSEXITCODE2(error, nd);
649	return (error);
650}
651
652/*
653 * Perform the NFSv4.1 destroy clientid.
654 */
655int
656nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
657{
658	struct nfsclient *clp;
659	struct nfsclienthashhead *hp;
660	int error = 0, i, igotlock;
661
662	if (nfsrvboottime != clientid.lval[0]) {
663		error = NFSERR_STALECLIENTID;
664		goto out;
665	}
666
667	/* Lock out other nfsd threads */
668	NFSLOCKV4ROOTMUTEX();
669	nfsv4_relref(&nfsv4rootfs_lock);
670	do {
671		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
672		    NFSV4ROOTLOCKMUTEXPTR, NULL);
673	} while (igotlock == 0);
674	NFSUNLOCKV4ROOTMUTEX();
675
676	hp = NFSCLIENTHASH(clientid);
677	LIST_FOREACH(clp, hp, lc_hash) {
678		if (clp->lc_clientid.lval[1] == clientid.lval[1])
679			break;
680	}
681	if (clp == NULL) {
682		NFSLOCKV4ROOTMUTEX();
683		nfsv4_unlock(&nfsv4rootfs_lock, 1);
684		NFSUNLOCKV4ROOTMUTEX();
685		/* Just return ok, since it is gone. */
686		goto out;
687	}
688
689	/* Scan for state on the clientid. */
690	for (i = 0; i < NFSSTATEHASHSIZE; i++)
691		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
692			NFSLOCKV4ROOTMUTEX();
693			nfsv4_unlock(&nfsv4rootfs_lock, 1);
694			NFSUNLOCKV4ROOTMUTEX();
695			error = NFSERR_CLIENTIDBUSY;
696			goto out;
697		}
698	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
699		NFSLOCKV4ROOTMUTEX();
700		nfsv4_unlock(&nfsv4rootfs_lock, 1);
701		NFSUNLOCKV4ROOTMUTEX();
702		error = NFSERR_CLIENTIDBUSY;
703		goto out;
704	}
705
706	/* Destroy the clientid and return ok. */
707	nfsrv_cleanclient(clp, p);
708	nfsrv_freedeleglist(&clp->lc_deleg);
709	nfsrv_freedeleglist(&clp->lc_olddeleg);
710	LIST_REMOVE(clp, lc_hash);
711	NFSLOCKV4ROOTMUTEX();
712	nfsv4_unlock(&nfsv4rootfs_lock, 1);
713	NFSUNLOCKV4ROOTMUTEX();
714	nfsrv_zapclient(clp, p);
715out:
716	NFSEXITCODE2(error, nd);
717	return (error);
718}
719
720/*
721 * Called from the new nfssvc syscall to admin revoke a clientid.
722 * Returns 0 for success, error otherwise.
723 */
724APPLESTATIC int
725nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
726{
727	struct nfsclient *clp = NULL;
728	int i, error = 0;
729	int gotit, igotlock;
730
731	/*
732	 * First, lock out the nfsd so that state won't change while the
733	 * revocation record is being written to the stable storage restart
734	 * file.
735	 */
736	NFSLOCKV4ROOTMUTEX();
737	do {
738		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
739		    NFSV4ROOTLOCKMUTEXPTR, NULL);
740	} while (!igotlock);
741	NFSUNLOCKV4ROOTMUTEX();
742
743	/*
744	 * Search for a match in the client list.
745	 */
746	gotit = i = 0;
747	while (i < NFSCLIENTHASHSIZE && !gotit) {
748	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
749		if (revokep->nclid_idlen == clp->lc_idlen &&
750		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
751			gotit = 1;
752			break;
753		}
754	    }
755	    i++;
756	}
757	if (!gotit) {
758		NFSLOCKV4ROOTMUTEX();
759		nfsv4_unlock(&nfsv4rootfs_lock, 0);
760		NFSUNLOCKV4ROOTMUTEX();
761		error = EPERM;
762		goto out;
763	}
764
765	/*
766	 * Now, write out the revocation record
767	 */
768	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
769	nfsrv_backupstable();
770
771	/*
772	 * and clear out the state, marking the clientid revoked.
773	 */
774	clp->lc_flags &= ~LCL_CALLBACKSON;
775	clp->lc_flags |= LCL_ADMINREVOKED;
776	nfsrv_cleanclient(clp, p);
777	nfsrv_freedeleglist(&clp->lc_deleg);
778	nfsrv_freedeleglist(&clp->lc_olddeleg);
779	NFSLOCKV4ROOTMUTEX();
780	nfsv4_unlock(&nfsv4rootfs_lock, 0);
781	NFSUNLOCKV4ROOTMUTEX();
782
783out:
784	NFSEXITCODE(error);
785	return (error);
786}
787
788/*
789 * Dump out stats for all clients. Called from nfssvc(2), that is used
790 * newnfsstats.
791 */
792APPLESTATIC void
793nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
794{
795	struct nfsclient *clp;
796	int i = 0, cnt = 0;
797
798	/*
799	 * First, get a reference on the nfsv4rootfs_lock so that an
800	 * exclusive lock cannot be acquired while dumping the clients.
801	 */
802	NFSLOCKV4ROOTMUTEX();
803	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
804	NFSUNLOCKV4ROOTMUTEX();
805	NFSLOCKSTATE();
806	/*
807	 * Rattle through the client lists until done.
808	 */
809	while (i < NFSCLIENTHASHSIZE && cnt < maxcnt) {
810	    clp = LIST_FIRST(&nfsclienthash[i]);
811	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
812		nfsrv_dumpaclient(clp, &dumpp[cnt]);
813		cnt++;
814		clp = LIST_NEXT(clp, lc_hash);
815	    }
816	    i++;
817	}
818	if (cnt < maxcnt)
819	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
820	NFSUNLOCKSTATE();
821	NFSLOCKV4ROOTMUTEX();
822	nfsv4_relref(&nfsv4rootfs_lock);
823	NFSUNLOCKV4ROOTMUTEX();
824}
825
826/*
827 * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
828 */
829static void
830nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
831{
832	struct nfsstate *stp, *openstp, *lckownstp;
833	struct nfslock *lop;
834	struct sockaddr *sad;
835	struct sockaddr_in *rad;
836	struct sockaddr_in6 *rad6;
837
838	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
839	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
840	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
841	dumpp->ndcl_flags = clp->lc_flags;
842	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
843	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
844	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
845	dumpp->ndcl_addrfam = sad->sa_family;
846	if (sad->sa_family == AF_INET) {
847		rad = (struct sockaddr_in *)sad;
848		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
849	} else {
850		rad6 = (struct sockaddr_in6 *)sad;
851		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
852	}
853
854	/*
855	 * Now, scan the state lists and total up the opens and locks.
856	 */
857	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
858	    dumpp->ndcl_nopenowners++;
859	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
860		dumpp->ndcl_nopens++;
861		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
862		    dumpp->ndcl_nlockowners++;
863		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
864			dumpp->ndcl_nlocks++;
865		    }
866		}
867	    }
868	}
869
870	/*
871	 * and the delegation lists.
872	 */
873	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
874	    dumpp->ndcl_ndelegs++;
875	}
876	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
877	    dumpp->ndcl_nolddelegs++;
878	}
879}
880
881/*
882 * Dump out lock stats for a file.
883 */
884APPLESTATIC void
885nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
886    NFSPROC_T *p)
887{
888	struct nfsstate *stp;
889	struct nfslock *lop;
890	int cnt = 0;
891	struct nfslockfile *lfp;
892	struct sockaddr *sad;
893	struct sockaddr_in *rad;
894	struct sockaddr_in6 *rad6;
895	int ret;
896	fhandle_t nfh;
897
898	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
899	/*
900	 * First, get a reference on the nfsv4rootfs_lock so that an
901	 * exclusive lock on it cannot be acquired while dumping the locks.
902	 */
903	NFSLOCKV4ROOTMUTEX();
904	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
905	NFSUNLOCKV4ROOTMUTEX();
906	NFSLOCKSTATE();
907	if (!ret)
908		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
909	if (ret) {
910		ldumpp[0].ndlck_clid.nclid_idlen = 0;
911		NFSUNLOCKSTATE();
912		NFSLOCKV4ROOTMUTEX();
913		nfsv4_relref(&nfsv4rootfs_lock);
914		NFSUNLOCKV4ROOTMUTEX();
915		return;
916	}
917
918	/*
919	 * For each open share on file, dump it out.
920	 */
921	stp = LIST_FIRST(&lfp->lf_open);
922	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
923		ldumpp[cnt].ndlck_flags = stp->ls_flags;
924		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
925		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
926		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
927		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
928		ldumpp[cnt].ndlck_owner.nclid_idlen =
929		    stp->ls_openowner->ls_ownerlen;
930		NFSBCOPY(stp->ls_openowner->ls_owner,
931		    ldumpp[cnt].ndlck_owner.nclid_id,
932		    stp->ls_openowner->ls_ownerlen);
933		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
934		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
935		    stp->ls_clp->lc_idlen);
936		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
937		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
938		if (sad->sa_family == AF_INET) {
939			rad = (struct sockaddr_in *)sad;
940			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
941		} else {
942			rad6 = (struct sockaddr_in6 *)sad;
943			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
944		}
945		stp = LIST_NEXT(stp, ls_file);
946		cnt++;
947	}
948
949	/*
950	 * and all locks.
951	 */
952	lop = LIST_FIRST(&lfp->lf_lock);
953	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
954		stp = lop->lo_stp;
955		ldumpp[cnt].ndlck_flags = lop->lo_flags;
956		ldumpp[cnt].ndlck_first = lop->lo_first;
957		ldumpp[cnt].ndlck_end = lop->lo_end;
958		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
959		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
960		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
961		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
962		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
963		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
964		    stp->ls_ownerlen);
965		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
966		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
967		    stp->ls_clp->lc_idlen);
968		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
969		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
970		if (sad->sa_family == AF_INET) {
971			rad = (struct sockaddr_in *)sad;
972			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
973		} else {
974			rad6 = (struct sockaddr_in6 *)sad;
975			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
976		}
977		lop = LIST_NEXT(lop, lo_lckfile);
978		cnt++;
979	}
980
981	/*
982	 * and the delegations.
983	 */
984	stp = LIST_FIRST(&lfp->lf_deleg);
985	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
986		ldumpp[cnt].ndlck_flags = stp->ls_flags;
987		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
988		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
989		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
990		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
991		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
992		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
993		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
994		    stp->ls_clp->lc_idlen);
995		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
996		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
997		if (sad->sa_family == AF_INET) {
998			rad = (struct sockaddr_in *)sad;
999			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1000		} else {
1001			rad6 = (struct sockaddr_in6 *)sad;
1002			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1003		}
1004		stp = LIST_NEXT(stp, ls_file);
1005		cnt++;
1006	}
1007
1008	/*
1009	 * If list isn't full, mark end of list by setting the client name
1010	 * to zero length.
1011	 */
1012	if (cnt < maxcnt)
1013		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1014	NFSUNLOCKSTATE();
1015	NFSLOCKV4ROOTMUTEX();
1016	nfsv4_relref(&nfsv4rootfs_lock);
1017	NFSUNLOCKV4ROOTMUTEX();
1018}
1019
1020/*
1021 * Server timer routine. It can scan any linked list, so long
1022 * as it holds the spin/mutex lock and there is no exclusive lock on
1023 * nfsv4rootfs_lock.
1024 * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1025 *  to do this from a callout, since the spin locks work. For
1026 *  Darwin, I'm not sure what will work correctly yet.)
1027 * Should be called once per second.
1028 */
1029APPLESTATIC void
1030nfsrv_servertimer(void)
1031{
1032	struct nfsclient *clp, *nclp;
1033	struct nfsstate *stp, *nstp;
1034	int got_ref, i;
1035
1036	/*
1037	 * Make sure nfsboottime is set. This is used by V3 as well
1038	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1039	 * only used by the V4 server for leases.
1040	 */
1041	if (nfsboottime.tv_sec == 0)
1042		NFSSETBOOTTIME(nfsboottime);
1043
1044	/*
1045	 * If server hasn't started yet, just return.
1046	 */
1047	NFSLOCKSTATE();
1048	if (nfsrv_stablefirst.nsf_eograce == 0) {
1049		NFSUNLOCKSTATE();
1050		return;
1051	}
1052	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1053		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1054		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1055			nfsrv_stablefirst.nsf_flags |=
1056			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1057		NFSUNLOCKSTATE();
1058		return;
1059	}
1060
1061	/*
1062	 * Try and get a reference count on the nfsv4rootfs_lock so that
1063	 * no nfsd thread can acquire an exclusive lock on it before this
1064	 * call is done. If it is already exclusively locked, just return.
1065	 */
1066	NFSLOCKV4ROOTMUTEX();
1067	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1068	NFSUNLOCKV4ROOTMUTEX();
1069	if (got_ref == 0) {
1070		NFSUNLOCKSTATE();
1071		return;
1072	}
1073
1074	/*
1075	 * For each client...
1076	 */
1077	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
1078	    clp = LIST_FIRST(&nfsclienthash[i]);
1079	    while (clp != LIST_END(&nfsclienthash[i])) {
1080		nclp = LIST_NEXT(clp, lc_hash);
1081		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1082		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1083			 && ((LIST_EMPTY(&clp->lc_deleg)
1084			      && LIST_EMPTY(&clp->lc_open)) ||
1085			     nfsrv_clients > nfsrv_clienthighwater)) ||
1086			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1087			(clp->lc_expiry < NFSD_MONOSEC &&
1088			 (nfsrv_openpluslock * 10 / 9) > NFSRV_V4STATELIMIT)) {
1089			/*
1090			 * Lease has expired several nfsrv_lease times ago:
1091			 * PLUS
1092			 *    - no state is associated with it
1093			 *    OR
1094			 *    - above high water mark for number of clients
1095			 *      (nfsrv_clienthighwater should be large enough
1096			 *       that this only occurs when clients fail to
1097			 *       use the same nfs_client_id4.id. Maybe somewhat
1098			 *       higher that the maximum number of clients that
1099			 *       will mount this server?)
1100			 * OR
1101			 * Lease has expired a very long time ago
1102			 * OR
1103			 * Lease has expired PLUS the number of opens + locks
1104			 * has exceeded 90% of capacity
1105			 *
1106			 * --> Mark for expiry. The actual expiry will be done
1107			 *     by an nfsd sometime soon.
1108			 */
1109			clp->lc_flags |= LCL_EXPIREIT;
1110			nfsrv_stablefirst.nsf_flags |=
1111			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1112		    } else {
1113			/*
1114			 * If there are no opens, increment no open tick cnt
1115			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1116			 * otherwise, if there is an open, reset no open time
1117			 * Hopefully, this will avoid excessive re-creation
1118			 * of open owners and subsequent open confirms.
1119			 */
1120			stp = LIST_FIRST(&clp->lc_open);
1121			while (stp != LIST_END(&clp->lc_open)) {
1122				nstp = LIST_NEXT(stp, ls_list);
1123				if (LIST_EMPTY(&stp->ls_open)) {
1124					stp->ls_noopens++;
1125					if (stp->ls_noopens > NFSNOOPEN ||
1126					    (nfsrv_openpluslock * 2) >
1127					    NFSRV_V4STATELIMIT)
1128						nfsrv_stablefirst.nsf_flags |=
1129							NFSNSF_NOOPENS;
1130				} else {
1131					stp->ls_noopens = 0;
1132				}
1133				stp = nstp;
1134			}
1135		    }
1136		}
1137		clp = nclp;
1138	    }
1139	}
1140	NFSUNLOCKSTATE();
1141	NFSLOCKV4ROOTMUTEX();
1142	nfsv4_relref(&nfsv4rootfs_lock);
1143	NFSUNLOCKV4ROOTMUTEX();
1144}
1145
1146/*
1147 * The following set of functions free up the various data structures.
1148 */
1149/*
1150 * Clear out all open/lock state related to this nfsclient.
1151 * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1152 * there are no other active nfsd threads.
1153 */
1154APPLESTATIC void
1155nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1156{
1157	struct nfsstate *stp, *nstp;
1158	struct nfsdsession *sep, *nsep;
1159
1160	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1161		nfsrv_freeopenowner(stp, 1, p);
1162	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1163		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1164			(void)nfsrv_freesession(sep, NULL);
1165}
1166
1167/*
1168 * Free a client that has been cleaned. It should also already have been
1169 * removed from the lists.
1170 * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1171 *  softclock interrupts are enabled.)
1172 */
1173APPLESTATIC void
1174nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1175{
1176
1177#ifdef notyet
1178	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1179	     (LCL_GSS | LCL_CALLBACKSON) &&
1180	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1181	    clp->lc_handlelen > 0) {
1182		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1183		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1184		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1185			NULL, 0, NULL, NULL, NULL, p);
1186	}
1187#endif
1188	newnfs_disconnect(&clp->lc_req);
1189	NFSSOCKADDRFREE(clp->lc_req.nr_nam);
1190	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1191	free((caddr_t)clp, M_NFSDCLIENT);
1192	NFSLOCKSTATE();
1193	newnfsstats.srvclients--;
1194	nfsrv_openpluslock--;
1195	nfsrv_clients--;
1196	NFSUNLOCKSTATE();
1197}
1198
1199/*
1200 * Free a list of delegation state structures.
1201 * (This function will also free all nfslockfile structures that no
1202 *  longer have associated state.)
1203 */
1204APPLESTATIC void
1205nfsrv_freedeleglist(struct nfsstatehead *sthp)
1206{
1207	struct nfsstate *stp, *nstp;
1208
1209	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1210		nfsrv_freedeleg(stp);
1211	}
1212	LIST_INIT(sthp);
1213}
1214
1215/*
1216 * Free up a delegation.
1217 */
1218static void
1219nfsrv_freedeleg(struct nfsstate *stp)
1220{
1221	struct nfslockfile *lfp;
1222
1223	LIST_REMOVE(stp, ls_hash);
1224	LIST_REMOVE(stp, ls_list);
1225	LIST_REMOVE(stp, ls_file);
1226	lfp = stp->ls_lfp;
1227	if (LIST_EMPTY(&lfp->lf_open) &&
1228	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1229	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1230	    lfp->lf_usecount == 0 &&
1231	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1232		nfsrv_freenfslockfile(lfp);
1233	FREE((caddr_t)stp, M_NFSDSTATE);
1234	newnfsstats.srvdelegates--;
1235	nfsrv_openpluslock--;
1236	nfsrv_delegatecnt--;
1237}
1238
1239/*
1240 * This function frees an open owner and all associated opens.
1241 */
1242static void
1243nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1244{
1245	struct nfsstate *nstp, *tstp;
1246
1247	LIST_REMOVE(stp, ls_list);
1248	/*
1249	 * Now, free all associated opens.
1250	 */
1251	nstp = LIST_FIRST(&stp->ls_open);
1252	while (nstp != LIST_END(&stp->ls_open)) {
1253		tstp = nstp;
1254		nstp = LIST_NEXT(nstp, ls_list);
1255		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1256	}
1257	if (stp->ls_op)
1258		nfsrvd_derefcache(stp->ls_op);
1259	FREE((caddr_t)stp, M_NFSDSTATE);
1260	newnfsstats.srvopenowners--;
1261	nfsrv_openpluslock--;
1262}
1263
1264/*
1265 * This function frees an open (nfsstate open structure) with all associated
1266 * lock_owners and locks. It also frees the nfslockfile structure iff there
1267 * are no other opens on the file.
1268 * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1269 */
1270static int
1271nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1272{
1273	struct nfsstate *nstp, *tstp;
1274	struct nfslockfile *lfp;
1275	int ret;
1276
1277	LIST_REMOVE(stp, ls_hash);
1278	LIST_REMOVE(stp, ls_list);
1279	LIST_REMOVE(stp, ls_file);
1280
1281	lfp = stp->ls_lfp;
1282	/*
1283	 * Now, free all lockowners associated with this open.
1284	 */
1285	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1286		nfsrv_freelockowner(tstp, vp, cansleep, p);
1287
1288	/*
1289	 * The nfslockfile is freed here if there are no locks
1290	 * associated with the open.
1291	 * If there are locks associated with the open, the
1292	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1293	 * Acquire the state mutex to avoid races with calls to
1294	 * nfsrv_getlockfile().
1295	 */
1296	if (cansleep != 0)
1297		NFSLOCKSTATE();
1298	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1299	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1300	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1301	    lfp->lf_usecount == 0 &&
1302	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1303		nfsrv_freenfslockfile(lfp);
1304		ret = 1;
1305	} else
1306		ret = 0;
1307	if (cansleep != 0)
1308		NFSUNLOCKSTATE();
1309	FREE((caddr_t)stp, M_NFSDSTATE);
1310	newnfsstats.srvopens--;
1311	nfsrv_openpluslock--;
1312	return (ret);
1313}
1314
1315/*
1316 * Frees a lockowner and all associated locks.
1317 */
1318static void
1319nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1320    NFSPROC_T *p)
1321{
1322
1323	LIST_REMOVE(stp, ls_hash);
1324	LIST_REMOVE(stp, ls_list);
1325	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1326	if (stp->ls_op)
1327		nfsrvd_derefcache(stp->ls_op);
1328	FREE((caddr_t)stp, M_NFSDSTATE);
1329	newnfsstats.srvlockowners--;
1330	nfsrv_openpluslock--;
1331}
1332
1333/*
1334 * Free all the nfs locks on a lockowner.
1335 */
1336static void
1337nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1338    NFSPROC_T *p)
1339{
1340	struct nfslock *lop, *nlop;
1341	struct nfsrollback *rlp, *nrlp;
1342	struct nfslockfile *lfp = NULL;
1343	int gottvp = 0;
1344	vnode_t tvp = NULL;
1345	uint64_t first, end;
1346
1347	lop = LIST_FIRST(&stp->ls_lock);
1348	while (lop != LIST_END(&stp->ls_lock)) {
1349		nlop = LIST_NEXT(lop, lo_lckowner);
1350		/*
1351		 * Since all locks should be for the same file, lfp should
1352		 * not change.
1353		 */
1354		if (lfp == NULL)
1355			lfp = lop->lo_lfp;
1356		else if (lfp != lop->lo_lfp)
1357			panic("allnfslocks");
1358		/*
1359		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1360		 * from the file handle. This only occurs when called from
1361		 * nfsrv_cleanclient().
1362		 */
1363		if (gottvp == 0) {
1364			if (nfsrv_dolocallocks == 0)
1365				tvp = NULL;
1366			else if (vp == NULL && cansleep != 0)
1367				tvp = nfsvno_getvp(&lfp->lf_fh);
1368			else
1369				tvp = vp;
1370			gottvp = 1;
1371		}
1372
1373		if (tvp != NULL) {
1374			if (cansleep == 0)
1375				panic("allnfs2");
1376			first = lop->lo_first;
1377			end = lop->lo_end;
1378			nfsrv_freenfslock(lop);
1379			nfsrv_localunlock(tvp, lfp, first, end, p);
1380			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1381			    nrlp)
1382				free(rlp, M_NFSDROLLBACK);
1383			LIST_INIT(&lfp->lf_rollback);
1384		} else
1385			nfsrv_freenfslock(lop);
1386		lop = nlop;
1387	}
1388	if (vp == NULL && tvp != NULL)
1389		vput(tvp);
1390}
1391
1392/*
1393 * Free an nfslock structure.
1394 */
1395static void
1396nfsrv_freenfslock(struct nfslock *lop)
1397{
1398
1399	if (lop->lo_lckfile.le_prev != NULL) {
1400		LIST_REMOVE(lop, lo_lckfile);
1401		newnfsstats.srvlocks--;
1402		nfsrv_openpluslock--;
1403	}
1404	LIST_REMOVE(lop, lo_lckowner);
1405	FREE((caddr_t)lop, M_NFSDLOCK);
1406}
1407
1408/*
1409 * This function frees an nfslockfile structure.
1410 */
1411static void
1412nfsrv_freenfslockfile(struct nfslockfile *lfp)
1413{
1414
1415	LIST_REMOVE(lfp, lf_hash);
1416	FREE((caddr_t)lfp, M_NFSDLOCKFILE);
1417}
1418
1419/*
1420 * This function looks up an nfsstate structure via stateid.
1421 */
1422static int
1423nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1424    struct nfsstate **stpp)
1425{
1426	struct nfsstate *stp;
1427	struct nfsstatehead *hp;
1428	int error = 0;
1429
1430	*stpp = NULL;
1431	hp = NFSSTATEHASH(clp, *stateidp);
1432	LIST_FOREACH(stp, hp, ls_hash) {
1433		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1434			NFSX_STATEIDOTHER))
1435			break;
1436	}
1437
1438	/*
1439	 * If no state id in list, return NFSERR_BADSTATEID.
1440	 */
1441	if (stp == LIST_END(hp)) {
1442		error = NFSERR_BADSTATEID;
1443		goto out;
1444	}
1445	*stpp = stp;
1446
1447out:
1448	NFSEXITCODE(error);
1449	return (error);
1450}
1451
1452/*
1453 * This function gets an nfsstate structure via owner string.
1454 */
1455static void
1456nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1457    struct nfsstate **stpp)
1458{
1459	struct nfsstate *stp;
1460
1461	*stpp = NULL;
1462	LIST_FOREACH(stp, hp, ls_list) {
1463		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1464		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1465			*stpp = stp;
1466			return;
1467		}
1468	}
1469}
1470
1471/*
1472 * Lock control function called to update lock status.
1473 * Returns 0 upon success, -1 if there is no lock and the flags indicate
1474 * that one isn't to be created and an NFSERR_xxx for other errors.
1475 * The structures new_stp and new_lop are passed in as pointers that should
1476 * be set to NULL if the structure is used and shouldn't be free'd.
1477 * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1478 * never used and can safely be allocated on the stack. For all other
1479 * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1480 * in case they are used.
1481 */
1482APPLESTATIC int
1483nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1484    struct nfslock **new_lopp, struct nfslockconflict *cfp,
1485    nfsquad_t clientid, nfsv4stateid_t *stateidp,
1486    __unused struct nfsexstuff *exp,
1487    struct nfsrv_descript *nd, NFSPROC_T *p)
1488{
1489	struct nfslock *lop;
1490	struct nfsstate *new_stp = *new_stpp;
1491	struct nfslock *new_lop = *new_lopp;
1492	struct nfsstate *tstp, *mystp, *nstp;
1493	int specialid = 0;
1494	struct nfslockfile *lfp;
1495	struct nfslock *other_lop = NULL;
1496	struct nfsstate *stp, *lckstp = NULL;
1497	struct nfsclient *clp = NULL;
1498	u_int32_t bits;
1499	int error = 0, haslock = 0, ret, reterr;
1500	int getlckret, delegation = 0, filestruct_locked;
1501	fhandle_t nfh;
1502	uint64_t first, end;
1503	uint32_t lock_flags;
1504
1505	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1506		/*
1507		 * Note the special cases of "all 1s" or "all 0s" stateids and
1508		 * let reads with all 1s go ahead.
1509		 */
1510		if (new_stp->ls_stateid.seqid == 0x0 &&
1511		    new_stp->ls_stateid.other[0] == 0x0 &&
1512		    new_stp->ls_stateid.other[1] == 0x0 &&
1513		    new_stp->ls_stateid.other[2] == 0x0)
1514			specialid = 1;
1515		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1516		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1517		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1518		    new_stp->ls_stateid.other[2] == 0xffffffff)
1519			specialid = 2;
1520	}
1521
1522	/*
1523	 * Check for restart conditions (client and server).
1524	 */
1525	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1526	    &new_stp->ls_stateid, specialid);
1527	if (error)
1528		goto out;
1529
1530	/*
1531	 * Check for state resource limit exceeded.
1532	 */
1533	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1534	    nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
1535		error = NFSERR_RESOURCE;
1536		goto out;
1537	}
1538
1539	/*
1540	 * For the lock case, get another nfslock structure,
1541	 * just in case we need it.
1542	 * Malloc now, before we start sifting through the linked lists,
1543	 * in case we have to wait for memory.
1544	 */
1545tryagain:
1546	if (new_stp->ls_flags & NFSLCK_LOCK)
1547		MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
1548		    M_NFSDLOCK, M_WAITOK);
1549	filestruct_locked = 0;
1550	reterr = 0;
1551	lfp = NULL;
1552
1553	/*
1554	 * Get the lockfile structure for CFH now, so we can do a sanity
1555	 * check against the stateid, before incrementing the seqid#, since
1556	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1557	 * shouldn't be incremented for this case.
1558	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1559	 * will be handled later.
1560	 * If we are doing Lock/LockU and local locking is enabled, sleep
1561	 * lock the nfslockfile structure.
1562	 */
1563	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1564	NFSLOCKSTATE();
1565	if (getlckret == 0) {
1566		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1567		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1568			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1569			    &lfp, &nfh, 1);
1570			if (getlckret == 0)
1571				filestruct_locked = 1;
1572		} else
1573			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1574			    &lfp, &nfh, 0);
1575	}
1576	if (getlckret != 0 && getlckret != -1)
1577		reterr = getlckret;
1578
1579	if (filestruct_locked != 0) {
1580		LIST_INIT(&lfp->lf_rollback);
1581		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1582			/*
1583			 * For local locking, do the advisory locking now, so
1584			 * that any conflict can be detected. A failure later
1585			 * can be rolled back locally. If an error is returned,
1586			 * struct nfslockfile has been unlocked and any local
1587			 * locking rolled back.
1588			 */
1589			NFSUNLOCKSTATE();
1590			reterr = nfsrv_locallock(vp, lfp,
1591			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1592			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1593			NFSLOCKSTATE();
1594		}
1595	}
1596
1597	if (specialid == 0) {
1598	    if (new_stp->ls_flags & NFSLCK_TEST) {
1599		/*
1600		 * RFC 3530 does not list LockT as an op that renews a
1601		 * lease, but the concensus seems to be that it is ok
1602		 * for a server to do so.
1603		 */
1604		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1605		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1606
1607		/*
1608		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1609		 * error returns for LockT, just go ahead and test for a lock,
1610		 * since there are no locks for this client, but other locks
1611		 * can conflict. (ie. same client will always be false)
1612		 */
1613		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1614		    error = 0;
1615		lckstp = new_stp;
1616	    } else {
1617	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1618		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1619	      if (error == 0)
1620		/*
1621		 * Look up the stateid
1622		 */
1623		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1624		  new_stp->ls_flags, &stp);
1625	      /*
1626	       * do some sanity checks for an unconfirmed open or a
1627	       * stateid that refers to the wrong file, for an open stateid
1628	       */
1629	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1630		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1631		   (getlckret == 0 && stp->ls_lfp != lfp)))
1632			error = NFSERR_BADSTATEID;
1633	      if (error == 0 &&
1634		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1635		  getlckret == 0 && stp->ls_lfp != lfp)
1636			error = NFSERR_BADSTATEID;
1637
1638	      /*
1639	       * If the lockowner stateid doesn't refer to the same file,
1640	       * I believe that is considered ok, since some clients will
1641	       * only create a single lockowner and use that for all locks
1642	       * on all files.
1643	       * For now, log it as a diagnostic, instead of considering it
1644	       * a BadStateid.
1645	       */
1646	      if (error == 0 && (stp->ls_flags &
1647		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1648		  getlckret == 0 && stp->ls_lfp != lfp) {
1649#ifdef DIAGNOSTIC
1650		  printf("Got a lock statid for different file open\n");
1651#endif
1652		  /*
1653		  error = NFSERR_BADSTATEID;
1654		  */
1655	      }
1656
1657	      if (error == 0) {
1658		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1659			/*
1660			 * If haslock set, we've already checked the seqid.
1661			 */
1662			if (!haslock) {
1663			    if (stp->ls_flags & NFSLCK_OPEN)
1664				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1665				    stp->ls_openowner, new_stp->ls_op);
1666			    else
1667				error = NFSERR_BADSTATEID;
1668			}
1669			if (!error)
1670			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1671			if (lckstp)
1672			    /*
1673			     * I believe this should be an error, but it
1674			     * isn't obvious what NFSERR_xxx would be
1675			     * appropriate, so I'll use NFSERR_INVAL for now.
1676			     */
1677			    error = NFSERR_INVAL;
1678			else
1679			    lckstp = new_stp;
1680		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1681			/*
1682			 * If haslock set, ditto above.
1683			 */
1684			if (!haslock) {
1685			    if (stp->ls_flags & NFSLCK_OPEN)
1686				error = NFSERR_BADSTATEID;
1687			    else
1688				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1689				    stp, new_stp->ls_op);
1690			}
1691			lckstp = stp;
1692		    } else {
1693			lckstp = stp;
1694		    }
1695	      }
1696	      /*
1697	       * If the seqid part of the stateid isn't the same, return
1698	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1699	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1700	       * nfsrv_returnoldstateid is set. (The concensus on the email
1701	       * list was that most clients would prefer to not receive
1702	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1703	       * is what will happen, so I use the nfsrv_returnoldstateid to
1704	       * allow for either server configuration.)
1705	       */
1706	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1707		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1708		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1709		    nfsrv_returnoldstateid)) ||
1710		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1711		    new_stp->ls_stateid.seqid != 0)))
1712		    error = NFSERR_OLDSTATEID;
1713	    }
1714	}
1715
1716	/*
1717	 * Now we can check for grace.
1718	 */
1719	if (!error)
1720		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1721	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1722		nfsrv_checkstable(clp))
1723		error = NFSERR_NOGRACE;
1724	/*
1725	 * If we successfully Reclaimed state, note that.
1726	 */
1727	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1728		nfsrv_markstable(clp);
1729
1730	/*
1731	 * At this point, either error == NFSERR_BADSTATEID or the
1732	 * seqid# has been updated, so we can return any error.
1733	 * If error == 0, there may be an error in:
1734	 *    nd_repstat - Set by the calling function.
1735	 *    reterr - Set above, if getting the nfslockfile structure
1736	 *       or acquiring the local lock failed.
1737	 *    (If both of these are set, nd_repstat should probably be
1738	 *     returned, since that error was detected before this
1739	 *     function call.)
1740	 */
1741	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1742		if (error == 0) {
1743			if (nd->nd_repstat != 0)
1744				error = nd->nd_repstat;
1745			else
1746				error = reterr;
1747		}
1748		if (filestruct_locked != 0) {
1749			/* Roll back local locks. */
1750			NFSUNLOCKSTATE();
1751			nfsrv_locallock_rollback(vp, lfp, p);
1752			NFSLOCKSTATE();
1753			nfsrv_unlocklf(lfp);
1754		}
1755		NFSUNLOCKSTATE();
1756		goto out;
1757	}
1758
1759	/*
1760	 * Check the nfsrv_getlockfile return.
1761	 * Returned -1 if no structure found.
1762	 */
1763	if (getlckret == -1) {
1764		error = NFSERR_EXPIRED;
1765		/*
1766		 * Called from lockt, so no lock is OK.
1767		 */
1768		if (new_stp->ls_flags & NFSLCK_TEST) {
1769			error = 0;
1770		} else if (new_stp->ls_flags &
1771		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1772			/*
1773			 * Called to check for a lock, OK if the stateid is all
1774			 * 1s or all 0s, but there should be an nfsstate
1775			 * otherwise.
1776			 * (ie. If there is no open, I'll assume no share
1777			 *  deny bits.)
1778			 */
1779			if (specialid)
1780				error = 0;
1781			else
1782				error = NFSERR_BADSTATEID;
1783		}
1784		NFSUNLOCKSTATE();
1785		goto out;
1786	}
1787
1788	/*
1789	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1790	 * For NFSLCK_CHECK, allow a read if write access is granted,
1791	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1792	 * which implies a conflicting deny can't exist.
1793	 */
1794	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1795	    /*
1796	     * Four kinds of state id:
1797	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1798	     * - stateid for an open
1799	     * - stateid for a delegation
1800	     * - stateid for a lock owner
1801	     */
1802	    if (!specialid) {
1803		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1804		    delegation = 1;
1805		    mystp = stp;
1806		    nfsrv_delaydelegtimeout(stp);
1807	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1808		    mystp = stp;
1809		} else {
1810		    mystp = stp->ls_openstp;
1811		}
1812		/*
1813		 * If locking or checking, require correct access
1814		 * bit set.
1815		 */
1816		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1817		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1818		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1819		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1820		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1821		     !(mystp->ls_flags & NFSLCK_READACCESS)) ||
1822		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1823		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1824		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1825			if (filestruct_locked != 0) {
1826				/* Roll back local locks. */
1827				NFSUNLOCKSTATE();
1828				nfsrv_locallock_rollback(vp, lfp, p);
1829				NFSLOCKSTATE();
1830				nfsrv_unlocklf(lfp);
1831			}
1832			NFSUNLOCKSTATE();
1833			error = NFSERR_OPENMODE;
1834			goto out;
1835		}
1836	    } else
1837		mystp = NULL;
1838	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1839		/*
1840		 * Check for a conflicting deny bit.
1841		 */
1842		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1843		    if (tstp != mystp) {
1844			bits = tstp->ls_flags;
1845			bits >>= NFSLCK_SHIFT;
1846			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1847			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1848				vp, p);
1849			    if (ret == 1) {
1850				/*
1851				* nfsrv_clientconflict unlocks state
1852				 * when it returns non-zero.
1853				 */
1854				lckstp = NULL;
1855				goto tryagain;
1856			    }
1857			    if (ret == 0)
1858				NFSUNLOCKSTATE();
1859			    if (ret == 2)
1860				error = NFSERR_PERM;
1861			    else
1862				error = NFSERR_OPENMODE;
1863			    goto out;
1864			}
1865		    }
1866		}
1867
1868		/* We're outta here */
1869		NFSUNLOCKSTATE();
1870		goto out;
1871	    }
1872	}
1873
1874	/*
1875	 * For setattr, just get rid of all the Delegations for other clients.
1876	 */
1877	if (new_stp->ls_flags & NFSLCK_SETATTR) {
1878		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1879		if (ret) {
1880			/*
1881			 * nfsrv_cleandeleg() unlocks state when it
1882			 * returns non-zero.
1883			 */
1884			if (ret == -1) {
1885				lckstp = NULL;
1886				goto tryagain;
1887			}
1888			error = ret;
1889			goto out;
1890		}
1891		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1892		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1893		     LIST_EMPTY(&lfp->lf_deleg))) {
1894			NFSUNLOCKSTATE();
1895			goto out;
1896		}
1897	}
1898
1899	/*
1900	 * Check for a conflicting delegation. If one is found, call
1901	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1902	 * been set yet, it will get the lock. Otherwise, it will recall
1903	 * the delegation. Then, we try try again...
1904	 * I currently believe the conflict algorithm to be:
1905	 * For Lock Ops (Lock/LockT/LockU)
1906	 * - there is a conflict iff a different client has a write delegation
1907	 * For Reading (Read Op)
1908	 * - there is a conflict iff a different client has a write delegation
1909	 *   (the specialids are always a different client)
1910	 * For Writing (Write/Setattr of size)
1911	 * - there is a conflict if a different client has any delegation
1912	 * - there is a conflict if the same client has a read delegation
1913	 *   (I don't understand why this isn't allowed, but that seems to be
1914	 *    the current concensus?)
1915	 */
1916	tstp = LIST_FIRST(&lfp->lf_deleg);
1917	while (tstp != LIST_END(&lfp->lf_deleg)) {
1918	    nstp = LIST_NEXT(tstp, ls_file);
1919	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
1920		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1921		  (new_lop->lo_flags & NFSLCK_READ))) &&
1922		  clp != tstp->ls_clp &&
1923		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
1924		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1925		   (new_lop->lo_flags & NFSLCK_WRITE) &&
1926		  (clp != tstp->ls_clp ||
1927		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
1928		if (filestruct_locked != 0) {
1929			/* Roll back local locks. */
1930			NFSUNLOCKSTATE();
1931			nfsrv_locallock_rollback(vp, lfp, p);
1932			NFSLOCKSTATE();
1933			nfsrv_unlocklf(lfp);
1934		}
1935		ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
1936		if (ret) {
1937		    /*
1938		     * nfsrv_delegconflict unlocks state when it
1939		     * returns non-zero, which it always does.
1940		     */
1941		    if (other_lop) {
1942			FREE((caddr_t)other_lop, M_NFSDLOCK);
1943			other_lop = NULL;
1944		    }
1945		    if (ret == -1) {
1946			lckstp = NULL;
1947			goto tryagain;
1948		    }
1949		    error = ret;
1950		    goto out;
1951		}
1952		/* Never gets here. */
1953	    }
1954	    tstp = nstp;
1955	}
1956
1957	/*
1958	 * Handle the unlock case by calling nfsrv_updatelock().
1959	 * (Should I have done some access checking above for unlock? For now,
1960	 *  just let it happen.)
1961	 */
1962	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
1963		first = new_lop->lo_first;
1964		end = new_lop->lo_end;
1965		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
1966		stateidp->seqid = ++(stp->ls_stateid.seqid);
1967		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
1968			stateidp->seqid = stp->ls_stateid.seqid = 1;
1969		stateidp->other[0] = stp->ls_stateid.other[0];
1970		stateidp->other[1] = stp->ls_stateid.other[1];
1971		stateidp->other[2] = stp->ls_stateid.other[2];
1972		if (filestruct_locked != 0) {
1973			NFSUNLOCKSTATE();
1974			/* Update the local locks. */
1975			nfsrv_localunlock(vp, lfp, first, end, p);
1976			NFSLOCKSTATE();
1977			nfsrv_unlocklf(lfp);
1978		}
1979		NFSUNLOCKSTATE();
1980		goto out;
1981	}
1982
1983	/*
1984	 * Search for a conflicting lock. A lock conflicts if:
1985	 * - the lock range overlaps and
1986	 * - at least one lock is a write lock and
1987	 * - it is not owned by the same lock owner
1988	 */
1989	if (!delegation) {
1990	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
1991	    if (new_lop->lo_end > lop->lo_first &&
1992		new_lop->lo_first < lop->lo_end &&
1993		(new_lop->lo_flags == NFSLCK_WRITE ||
1994		 lop->lo_flags == NFSLCK_WRITE) &&
1995		lckstp != lop->lo_stp &&
1996		(clp != lop->lo_stp->ls_clp ||
1997		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
1998		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
1999		    lckstp->ls_ownerlen))) {
2000		if (other_lop) {
2001		    FREE((caddr_t)other_lop, M_NFSDLOCK);
2002		    other_lop = NULL;
2003		}
2004		ret = nfsrv_clientconflict(lop->lo_stp->ls_clp,&haslock,vp,p);
2005		if (ret == 1) {
2006		    if (filestruct_locked != 0) {
2007			/* Roll back local locks. */
2008			nfsrv_locallock_rollback(vp, lfp, p);
2009			NFSLOCKSTATE();
2010			nfsrv_unlocklf(lfp);
2011			NFSUNLOCKSTATE();
2012		    }
2013		    /*
2014		     * nfsrv_clientconflict() unlocks state when it
2015		     * returns non-zero.
2016		     */
2017		    lckstp = NULL;
2018		    goto tryagain;
2019		}
2020		/*
2021		 * Found a conflicting lock, so record the conflict and
2022		 * return the error.
2023		 */
2024		if (cfp != NULL && ret == 0) {
2025		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2026		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2027		    cfp->cl_first = lop->lo_first;
2028		    cfp->cl_end = lop->lo_end;
2029		    cfp->cl_flags = lop->lo_flags;
2030		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2031		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2032			cfp->cl_ownerlen);
2033		}
2034		if (ret == 2)
2035		    error = NFSERR_PERM;
2036		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2037		    error = NFSERR_RECLAIMCONFLICT;
2038		else if (new_stp->ls_flags & NFSLCK_CHECK)
2039		    error = NFSERR_LOCKED;
2040		else
2041		    error = NFSERR_DENIED;
2042		if (filestruct_locked != 0 && ret == 0) {
2043			/* Roll back local locks. */
2044			NFSUNLOCKSTATE();
2045			nfsrv_locallock_rollback(vp, lfp, p);
2046			NFSLOCKSTATE();
2047			nfsrv_unlocklf(lfp);
2048		}
2049		if (ret == 0)
2050			NFSUNLOCKSTATE();
2051		goto out;
2052	    }
2053	  }
2054	}
2055
2056	/*
2057	 * We only get here if there was no lock that conflicted.
2058	 */
2059	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2060		NFSUNLOCKSTATE();
2061		goto out;
2062	}
2063
2064	/*
2065	 * We only get here when we are creating or modifying a lock.
2066	 * There are two variants:
2067	 * - exist_lock_owner where lock_owner exists
2068	 * - open_to_lock_owner with new lock_owner
2069	 */
2070	first = new_lop->lo_first;
2071	end = new_lop->lo_end;
2072	lock_flags = new_lop->lo_flags;
2073	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2074		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2075		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2076		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2077			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2078		stateidp->other[0] = lckstp->ls_stateid.other[0];
2079		stateidp->other[1] = lckstp->ls_stateid.other[1];
2080		stateidp->other[2] = lckstp->ls_stateid.other[2];
2081	} else {
2082		/*
2083		 * The new open_to_lock_owner case.
2084		 * Link the new nfsstate into the lists.
2085		 */
2086		new_stp->ls_seq = new_stp->ls_opentolockseq;
2087		nfsrvd_refcache(new_stp->ls_op);
2088		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2089		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2090		    clp->lc_clientid.lval[0];
2091		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2092		    clp->lc_clientid.lval[1];
2093		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2094		    nfsrv_nextstateindex(clp);
2095		new_stp->ls_clp = clp;
2096		LIST_INIT(&new_stp->ls_lock);
2097		new_stp->ls_openstp = stp;
2098		new_stp->ls_lfp = lfp;
2099		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2100		    lfp);
2101		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2102		    new_stp, ls_hash);
2103		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2104		*new_lopp = NULL;
2105		*new_stpp = NULL;
2106		newnfsstats.srvlockowners++;
2107		nfsrv_openpluslock++;
2108	}
2109	if (filestruct_locked != 0) {
2110		NFSUNLOCKSTATE();
2111		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2112		NFSLOCKSTATE();
2113		nfsrv_unlocklf(lfp);
2114	}
2115	NFSUNLOCKSTATE();
2116
2117out:
2118	if (haslock) {
2119		NFSLOCKV4ROOTMUTEX();
2120		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2121		NFSUNLOCKV4ROOTMUTEX();
2122	}
2123	if (other_lop)
2124		FREE((caddr_t)other_lop, M_NFSDLOCK);
2125	NFSEXITCODE2(error, nd);
2126	return (error);
2127}
2128
2129/*
2130 * Check for state errors for Open.
2131 * repstat is passed back out as an error if more critical errors
2132 * are not detected.
2133 */
2134APPLESTATIC int
2135nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2136    struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2137    NFSPROC_T *p, int repstat)
2138{
2139	struct nfsstate *stp, *nstp;
2140	struct nfsclient *clp;
2141	struct nfsstate *ownerstp;
2142	struct nfslockfile *lfp, *new_lfp;
2143	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2144
2145	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2146		readonly = 1;
2147	/*
2148	 * Check for restart conditions (client and server).
2149	 */
2150	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2151		&new_stp->ls_stateid, 0);
2152	if (error)
2153		goto out;
2154
2155	/*
2156	 * Check for state resource limit exceeded.
2157	 * Technically this should be SMP protected, but the worst
2158	 * case error is "out by one or two" on the count when it
2159	 * returns NFSERR_RESOURCE and the limit is just a rather
2160	 * arbitrary high water mark, so no harm is done.
2161	 */
2162	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
2163		error = NFSERR_RESOURCE;
2164		goto out;
2165	}
2166
2167tryagain:
2168	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2169	    M_NFSDLOCKFILE, M_WAITOK);
2170	if (vp)
2171		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2172		    NULL, p);
2173	NFSLOCKSTATE();
2174	/*
2175	 * Get the nfsclient structure.
2176	 */
2177	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2178	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2179
2180	/*
2181	 * Look up the open owner. See if it needs confirmation and
2182	 * check the seq#, as required.
2183	 */
2184	if (!error)
2185		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2186
2187	if (!error && ownerstp) {
2188		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2189		    new_stp->ls_op);
2190		/*
2191		 * If the OpenOwner hasn't been confirmed, assume the
2192		 * old one was a replay and this one is ok.
2193		 * See: RFC3530 Sec. 14.2.18.
2194		 */
2195		if (error == NFSERR_BADSEQID &&
2196		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2197			error = 0;
2198	}
2199
2200	/*
2201	 * Check for grace.
2202	 */
2203	if (!error)
2204		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2205	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2206		nfsrv_checkstable(clp))
2207		error = NFSERR_NOGRACE;
2208
2209	/*
2210	 * If none of the above errors occurred, let repstat be
2211	 * returned.
2212	 */
2213	if (repstat && !error)
2214		error = repstat;
2215	if (error) {
2216		NFSUNLOCKSTATE();
2217		if (haslock) {
2218			NFSLOCKV4ROOTMUTEX();
2219			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2220			NFSUNLOCKV4ROOTMUTEX();
2221		}
2222		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2223		goto out;
2224	}
2225
2226	/*
2227	 * If vp == NULL, the file doesn't exist yet, so return ok.
2228	 * (This always happens on the first pass, so haslock must be 0.)
2229	 */
2230	if (vp == NULL) {
2231		NFSUNLOCKSTATE();
2232		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2233		goto out;
2234	}
2235
2236	/*
2237	 * Get the structure for the underlying file.
2238	 */
2239	if (getfhret)
2240		error = getfhret;
2241	else
2242		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2243		    NULL, 0);
2244	if (new_lfp)
2245		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2246	if (error) {
2247		NFSUNLOCKSTATE();
2248		if (haslock) {
2249			NFSLOCKV4ROOTMUTEX();
2250			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2251			NFSUNLOCKV4ROOTMUTEX();
2252		}
2253		goto out;
2254	}
2255
2256	/*
2257	 * Search for a conflicting open/share.
2258	 */
2259	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2260	    /*
2261	     * For Delegate_Cur, search for the matching Delegation,
2262	     * which indicates no conflict.
2263	     * An old delegation should have been recovered by the
2264	     * client doing a Claim_DELEGATE_Prev, so I won't let
2265	     * it match and return NFSERR_EXPIRED. Should I let it
2266	     * match?
2267	     */
2268	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2269		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2270		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2271		    stateidp->seqid == 0) ||
2272		    stateidp->seqid == stp->ls_stateid.seqid) &&
2273		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2274			  NFSX_STATEIDOTHER))
2275			break;
2276	    }
2277	    if (stp == LIST_END(&lfp->lf_deleg) ||
2278		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2279		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2280		NFSUNLOCKSTATE();
2281		if (haslock) {
2282			NFSLOCKV4ROOTMUTEX();
2283			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2284			NFSUNLOCKV4ROOTMUTEX();
2285		}
2286		error = NFSERR_EXPIRED;
2287		goto out;
2288	    }
2289	}
2290
2291	/*
2292	 * Check for access/deny bit conflicts. I check for the same
2293	 * owner as well, in case the client didn't bother.
2294	 */
2295	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2296		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2297		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2298		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2299		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2300		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2301			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2302			if (ret == 1) {
2303				/*
2304				 * nfsrv_clientconflict() unlocks
2305				 * state when it returns non-zero.
2306				 */
2307				goto tryagain;
2308			}
2309			if (ret == 2)
2310				error = NFSERR_PERM;
2311			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2312				error = NFSERR_RECLAIMCONFLICT;
2313			else
2314				error = NFSERR_SHAREDENIED;
2315			if (ret == 0)
2316				NFSUNLOCKSTATE();
2317			if (haslock) {
2318				NFSLOCKV4ROOTMUTEX();
2319				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2320				NFSUNLOCKV4ROOTMUTEX();
2321			}
2322			goto out;
2323		}
2324	}
2325
2326	/*
2327	 * Check for a conflicting delegation. If one is found, call
2328	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2329	 * been set yet, it will get the lock. Otherwise, it will recall
2330	 * the delegation. Then, we try try again...
2331	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2332	 *  isn't a conflict.)
2333	 * I currently believe the conflict algorithm to be:
2334	 * For Open with Read Access and Deny None
2335	 * - there is a conflict iff a different client has a write delegation
2336	 * For Open with other Write Access or any Deny except None
2337	 * - there is a conflict if a different client has any delegation
2338	 * - there is a conflict if the same client has a read delegation
2339	 *   (The current concensus is that this last case should be
2340	 *    considered a conflict since the client with a read delegation
2341	 *    could have done an Open with ReadAccess and WriteDeny
2342	 *    locally and then not have checked for the WriteDeny.)
2343	 * Don't check for a Reclaim, since that will be dealt with
2344	 * by nfsrv_openctrl().
2345	 */
2346	if (!(new_stp->ls_flags &
2347		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2348	    stp = LIST_FIRST(&lfp->lf_deleg);
2349	    while (stp != LIST_END(&lfp->lf_deleg)) {
2350		nstp = LIST_NEXT(stp, ls_file);
2351		if ((readonly && stp->ls_clp != clp &&
2352		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2353		    (!readonly && (stp->ls_clp != clp ||
2354		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2355			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2356			if (ret) {
2357			    /*
2358			     * nfsrv_delegconflict() unlocks state
2359			     * when it returns non-zero.
2360			     */
2361			    if (ret == -1)
2362				goto tryagain;
2363			    error = ret;
2364			    goto out;
2365			}
2366		}
2367		stp = nstp;
2368	    }
2369	}
2370	NFSUNLOCKSTATE();
2371	if (haslock) {
2372		NFSLOCKV4ROOTMUTEX();
2373		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2374		NFSUNLOCKV4ROOTMUTEX();
2375	}
2376
2377out:
2378	NFSEXITCODE2(error, nd);
2379	return (error);
2380}
2381
2382/*
2383 * Open control function to create/update open state for an open.
2384 */
2385APPLESTATIC int
2386nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2387    struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2388    nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2389    NFSPROC_T *p, u_quad_t filerev)
2390{
2391	struct nfsstate *new_stp = *new_stpp;
2392	struct nfsstate *stp, *nstp;
2393	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2394	struct nfslockfile *lfp, *new_lfp;
2395	struct nfsclient *clp;
2396	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2397	int readonly = 0, cbret = 1, getfhret = 0;
2398
2399	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2400		readonly = 1;
2401	/*
2402	 * Check for restart conditions (client and server).
2403	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2404	 * If an error does show up, return NFSERR_EXPIRED, since the
2405	 * the seqid# has already been incremented.
2406	 */
2407	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2408	    &new_stp->ls_stateid, 0);
2409	if (error) {
2410		printf("Nfsd: openctrl unexpected restart err=%d\n",
2411		    error);
2412		error = NFSERR_EXPIRED;
2413		goto out;
2414	}
2415
2416tryagain:
2417	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2418	    M_NFSDLOCKFILE, M_WAITOK);
2419	MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2420	    M_NFSDSTATE, M_WAITOK);
2421	MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2422	    M_NFSDSTATE, M_WAITOK);
2423	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2424	    NULL, p);
2425	NFSLOCKSTATE();
2426	/*
2427	 * Get the client structure. Since the linked lists could be changed
2428	 * by other nfsd processes if this process does a tsleep(), one of
2429	 * two things must be done.
2430	 * 1 - don't tsleep()
2431	 * or
2432	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2433	 *     before using the lists, since this lock stops the other
2434	 *     nfsd. This should only be used for rare cases, since it
2435	 *     essentially single threads the nfsd.
2436	 *     At this time, it is only done for cases where the stable
2437	 *     storage file must be written prior to completion of state
2438	 *     expiration.
2439	 */
2440	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2441	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2442	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2443	    clp->lc_program) {
2444		/*
2445		 * This happens on the first open for a client
2446		 * that supports callbacks.
2447		 */
2448		NFSUNLOCKSTATE();
2449		/*
2450		 * Although nfsrv_docallback() will sleep, clp won't
2451		 * go away, since they are only removed when the
2452		 * nfsv4_lock() has blocked the nfsd threads. The
2453		 * fields in clp can change, but having multiple
2454		 * threads do this Null callback RPC should be
2455		 * harmless.
2456		 */
2457		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2458		    NULL, 0, NULL, NULL, NULL, p);
2459		NFSLOCKSTATE();
2460		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2461		if (!cbret)
2462			clp->lc_flags |= LCL_CALLBACKSON;
2463	}
2464
2465	/*
2466	 * Look up the open owner. See if it needs confirmation and
2467	 * check the seq#, as required.
2468	 */
2469	if (!error)
2470		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2471
2472	if (error) {
2473		NFSUNLOCKSTATE();
2474		printf("Nfsd: openctrl unexpected state err=%d\n",
2475			error);
2476		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2477		free((caddr_t)new_open, M_NFSDSTATE);
2478		free((caddr_t)new_deleg, M_NFSDSTATE);
2479		if (haslock) {
2480			NFSLOCKV4ROOTMUTEX();
2481			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2482			NFSUNLOCKV4ROOTMUTEX();
2483		}
2484		error = NFSERR_EXPIRED;
2485		goto out;
2486	}
2487
2488	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2489		nfsrv_markstable(clp);
2490
2491	/*
2492	 * Get the structure for the underlying file.
2493	 */
2494	if (getfhret)
2495		error = getfhret;
2496	else
2497		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2498		    NULL, 0);
2499	if (new_lfp)
2500		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2501	if (error) {
2502		NFSUNLOCKSTATE();
2503		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2504		    error);
2505		free((caddr_t)new_open, M_NFSDSTATE);
2506		free((caddr_t)new_deleg, M_NFSDSTATE);
2507		if (haslock) {
2508			NFSLOCKV4ROOTMUTEX();
2509			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2510			NFSUNLOCKV4ROOTMUTEX();
2511		}
2512		goto out;
2513	}
2514
2515	/*
2516	 * Search for a conflicting open/share.
2517	 */
2518	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2519	    /*
2520	     * For Delegate_Cur, search for the matching Delegation,
2521	     * which indicates no conflict.
2522	     * An old delegation should have been recovered by the
2523	     * client doing a Claim_DELEGATE_Prev, so I won't let
2524	     * it match and return NFSERR_EXPIRED. Should I let it
2525	     * match?
2526	     */
2527	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2528		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2529		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2530		    stateidp->seqid == 0) ||
2531		    stateidp->seqid == stp->ls_stateid.seqid) &&
2532		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2533			NFSX_STATEIDOTHER))
2534			break;
2535	    }
2536	    if (stp == LIST_END(&lfp->lf_deleg) ||
2537		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2538		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2539		NFSUNLOCKSTATE();
2540		printf("Nfsd openctrl unexpected expiry\n");
2541		free((caddr_t)new_open, M_NFSDSTATE);
2542		free((caddr_t)new_deleg, M_NFSDSTATE);
2543		if (haslock) {
2544			NFSLOCKV4ROOTMUTEX();
2545			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2546			NFSUNLOCKV4ROOTMUTEX();
2547		}
2548		error = NFSERR_EXPIRED;
2549		goto out;
2550	    }
2551
2552	    /*
2553	     * Don't issue a Delegation, since one already exists and
2554	     * delay delegation timeout, as required.
2555	     */
2556	    delegate = 0;
2557	    nfsrv_delaydelegtimeout(stp);
2558	}
2559
2560	/*
2561	 * Check for access/deny bit conflicts. I also check for the
2562	 * same owner, since the client might not have bothered to check.
2563	 * Also, note an open for the same file and owner, if found,
2564	 * which is all we do here for Delegate_Cur, since conflict
2565	 * checking is already done.
2566	 */
2567	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2568		if (ownerstp && stp->ls_openowner == ownerstp)
2569			openstp = stp;
2570		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2571		    /*
2572		     * If another client has the file open, the only
2573		     * delegation that can be issued is a Read delegation
2574		     * and only if it is a Read open with Deny none.
2575		     */
2576		    if (clp != stp->ls_clp) {
2577			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2578			    NFSLCK_READACCESS)
2579			    writedeleg = 0;
2580			else
2581			    delegate = 0;
2582		    }
2583		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2584		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2585		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2586		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2587			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2588			if (ret == 1) {
2589				/*
2590				 * nfsrv_clientconflict() unlocks state
2591				 * when it returns non-zero.
2592				 */
2593				free((caddr_t)new_open, M_NFSDSTATE);
2594				free((caddr_t)new_deleg, M_NFSDSTATE);
2595				openstp = NULL;
2596				goto tryagain;
2597			}
2598			if (ret == 2)
2599				error = NFSERR_PERM;
2600			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2601				error = NFSERR_RECLAIMCONFLICT;
2602			else
2603				error = NFSERR_SHAREDENIED;
2604			if (ret == 0)
2605				NFSUNLOCKSTATE();
2606			if (haslock) {
2607				NFSLOCKV4ROOTMUTEX();
2608				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2609				NFSUNLOCKV4ROOTMUTEX();
2610			}
2611			free((caddr_t)new_open, M_NFSDSTATE);
2612			free((caddr_t)new_deleg, M_NFSDSTATE);
2613			printf("nfsd openctrl unexpected client cnfl\n");
2614			goto out;
2615		    }
2616		}
2617	}
2618
2619	/*
2620	 * Check for a conflicting delegation. If one is found, call
2621	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2622	 * been set yet, it will get the lock. Otherwise, it will recall
2623	 * the delegation. Then, we try try again...
2624	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2625	 *  isn't a conflict.)
2626	 * I currently believe the conflict algorithm to be:
2627	 * For Open with Read Access and Deny None
2628	 * - there is a conflict iff a different client has a write delegation
2629	 * For Open with other Write Access or any Deny except None
2630	 * - there is a conflict if a different client has any delegation
2631	 * - there is a conflict if the same client has a read delegation
2632	 *   (The current concensus is that this last case should be
2633	 *    considered a conflict since the client with a read delegation
2634	 *    could have done an Open with ReadAccess and WriteDeny
2635	 *    locally and then not have checked for the WriteDeny.)
2636	 */
2637	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2638	    stp = LIST_FIRST(&lfp->lf_deleg);
2639	    while (stp != LIST_END(&lfp->lf_deleg)) {
2640		nstp = LIST_NEXT(stp, ls_file);
2641		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2642			writedeleg = 0;
2643		else
2644			delegate = 0;
2645		if ((readonly && stp->ls_clp != clp &&
2646		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2647		    (!readonly && (stp->ls_clp != clp ||
2648		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2649		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2650			delegate = 2;
2651		    } else {
2652			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2653			if (ret) {
2654			    /*
2655			     * nfsrv_delegconflict() unlocks state
2656			     * when it returns non-zero.
2657			     */
2658			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2659			    free((caddr_t)new_open, M_NFSDSTATE);
2660			    free((caddr_t)new_deleg, M_NFSDSTATE);
2661			    if (ret == -1) {
2662				openstp = NULL;
2663				goto tryagain;
2664			    }
2665			    error = ret;
2666			    goto out;
2667			}
2668		    }
2669		}
2670		stp = nstp;
2671	    }
2672	}
2673
2674	/*
2675	 * We only get here if there was no open that conflicted.
2676	 * If an open for the owner exists, or in the access/deny bits.
2677	 * Otherwise it is a new open. If the open_owner hasn't been
2678	 * confirmed, replace the open with the new one needing confirmation,
2679	 * otherwise add the open.
2680	 */
2681	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2682	    /*
2683	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2684	     * a match. If found, just move the old delegation to the current
2685	     * delegation list and issue open. If not found, return
2686	     * NFSERR_EXPIRED.
2687	     */
2688	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2689		if (stp->ls_lfp == lfp) {
2690		    /* Found it */
2691		    if (stp->ls_clp != clp)
2692			panic("olddeleg clp");
2693		    LIST_REMOVE(stp, ls_list);
2694		    LIST_REMOVE(stp, ls_hash);
2695		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2696		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2697		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2698			clp->lc_clientid.lval[0];
2699		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2700			clp->lc_clientid.lval[1];
2701		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2702			nfsrv_nextstateindex(clp);
2703		    stp->ls_compref = nd->nd_compref;
2704		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2705		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2706			stp->ls_stateid), stp, ls_hash);
2707		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2708			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2709		    else
2710			*rflagsp |= NFSV4OPEN_READDELEGATE;
2711		    clp->lc_delegtime = NFSD_MONOSEC +
2712			nfsrv_lease + NFSRV_LEASEDELTA;
2713
2714		    /*
2715		     * Now, do the associated open.
2716		     */
2717		    new_open->ls_stateid.seqid = 1;
2718		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2719		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2720		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2721		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2722			NFSLCK_OPEN;
2723		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2724			new_open->ls_flags |= (NFSLCK_READACCESS |
2725			    NFSLCK_WRITEACCESS);
2726		    else
2727			new_open->ls_flags |= NFSLCK_READACCESS;
2728		    new_open->ls_uid = new_stp->ls_uid;
2729		    new_open->ls_lfp = lfp;
2730		    new_open->ls_clp = clp;
2731		    LIST_INIT(&new_open->ls_open);
2732		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2733		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2734			new_open, ls_hash);
2735		    /*
2736		     * and handle the open owner
2737		     */
2738		    if (ownerstp) {
2739			new_open->ls_openowner = ownerstp;
2740			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2741		    } else {
2742			new_open->ls_openowner = new_stp;
2743			new_stp->ls_flags = 0;
2744			nfsrvd_refcache(new_stp->ls_op);
2745			new_stp->ls_noopens = 0;
2746			LIST_INIT(&new_stp->ls_open);
2747			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2748			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2749			*new_stpp = NULL;
2750			newnfsstats.srvopenowners++;
2751			nfsrv_openpluslock++;
2752		    }
2753		    openstp = new_open;
2754		    new_open = NULL;
2755		    newnfsstats.srvopens++;
2756		    nfsrv_openpluslock++;
2757		    break;
2758		}
2759	    }
2760	    if (stp == LIST_END(&clp->lc_olddeleg))
2761		error = NFSERR_EXPIRED;
2762	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2763	    /*
2764	     * Scan to see that no delegation for this client and file
2765	     * doesn't already exist.
2766	     * There also shouldn't yet be an Open for this file and
2767	     * openowner.
2768	     */
2769	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2770		if (stp->ls_clp == clp)
2771		    break;
2772	    }
2773	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2774		/*
2775		 * This is the Claim_Previous case with a delegation
2776		 * type != Delegate_None.
2777		 */
2778		/*
2779		 * First, add the delegation. (Although we must issue the
2780		 * delegation, we can also ask for an immediate return.)
2781		 */
2782		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2783		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2784		    clp->lc_clientid.lval[0];
2785		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2786		    clp->lc_clientid.lval[1];
2787		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2788		    nfsrv_nextstateindex(clp);
2789		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2790		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2791			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2792		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2793		} else {
2794		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2795			NFSLCK_READACCESS);
2796		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2797		}
2798		new_deleg->ls_uid = new_stp->ls_uid;
2799		new_deleg->ls_lfp = lfp;
2800		new_deleg->ls_clp = clp;
2801		new_deleg->ls_filerev = filerev;
2802		new_deleg->ls_compref = nd->nd_compref;
2803		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2804		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2805		    new_deleg->ls_stateid), new_deleg, ls_hash);
2806		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2807		new_deleg = NULL;
2808		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2809		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2810		     LCL_CALLBACKSON ||
2811		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2812		    !NFSVNO_DELEGOK(vp))
2813		    *rflagsp |= NFSV4OPEN_RECALL;
2814		newnfsstats.srvdelegates++;
2815		nfsrv_openpluslock++;
2816		nfsrv_delegatecnt++;
2817
2818		/*
2819		 * Now, do the associated open.
2820		 */
2821		new_open->ls_stateid.seqid = 1;
2822		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2823		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2824		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2825		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2826		    NFSLCK_OPEN;
2827		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2828			new_open->ls_flags |= (NFSLCK_READACCESS |
2829			    NFSLCK_WRITEACCESS);
2830		else
2831			new_open->ls_flags |= NFSLCK_READACCESS;
2832		new_open->ls_uid = new_stp->ls_uid;
2833		new_open->ls_lfp = lfp;
2834		new_open->ls_clp = clp;
2835		LIST_INIT(&new_open->ls_open);
2836		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2837		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2838		   new_open, ls_hash);
2839		/*
2840		 * and handle the open owner
2841		 */
2842		if (ownerstp) {
2843		    new_open->ls_openowner = ownerstp;
2844		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2845		} else {
2846		    new_open->ls_openowner = new_stp;
2847		    new_stp->ls_flags = 0;
2848		    nfsrvd_refcache(new_stp->ls_op);
2849		    new_stp->ls_noopens = 0;
2850		    LIST_INIT(&new_stp->ls_open);
2851		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2852		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2853		    *new_stpp = NULL;
2854		    newnfsstats.srvopenowners++;
2855		    nfsrv_openpluslock++;
2856		}
2857		openstp = new_open;
2858		new_open = NULL;
2859		newnfsstats.srvopens++;
2860		nfsrv_openpluslock++;
2861	    } else {
2862		error = NFSERR_RECLAIMCONFLICT;
2863	    }
2864	} else if (ownerstp) {
2865		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2866		    /* Replace the open */
2867		    if (ownerstp->ls_op)
2868			nfsrvd_derefcache(ownerstp->ls_op);
2869		    ownerstp->ls_op = new_stp->ls_op;
2870		    nfsrvd_refcache(ownerstp->ls_op);
2871		    ownerstp->ls_seq = new_stp->ls_seq;
2872		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2873		    stp = LIST_FIRST(&ownerstp->ls_open);
2874		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2875			NFSLCK_OPEN;
2876		    stp->ls_stateid.seqid = 1;
2877		    stp->ls_uid = new_stp->ls_uid;
2878		    if (lfp != stp->ls_lfp) {
2879			LIST_REMOVE(stp, ls_file);
2880			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
2881			stp->ls_lfp = lfp;
2882		    }
2883		    openstp = stp;
2884		} else if (openstp) {
2885		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
2886		    openstp->ls_stateid.seqid++;
2887		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
2888			openstp->ls_stateid.seqid == 0)
2889			openstp->ls_stateid.seqid = 1;
2890
2891		    /*
2892		     * This is where we can choose to issue a delegation.
2893		     */
2894		    if (delegate == 0 || writedeleg == 0 ||
2895			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
2896			nfsrv_writedelegifpos == 0) ||
2897			!NFSVNO_DELEGOK(vp) ||
2898			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
2899			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2900			 LCL_CALLBACKSON)
2901			*rflagsp |= NFSV4OPEN_WDCONTENTION;
2902		    else if (nfsrv_issuedelegs == 0 ||
2903			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
2904			*rflagsp |= NFSV4OPEN_WDRESOURCE;
2905		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
2906			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
2907		    else {
2908			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2909			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2910			    = clp->lc_clientid.lval[0];
2911			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2912			    = clp->lc_clientid.lval[1];
2913			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2914			    = nfsrv_nextstateindex(clp);
2915			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2916			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2917			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2918			new_deleg->ls_uid = new_stp->ls_uid;
2919			new_deleg->ls_lfp = lfp;
2920			new_deleg->ls_clp = clp;
2921			new_deleg->ls_filerev = filerev;
2922			new_deleg->ls_compref = nd->nd_compref;
2923			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2924			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2925			    new_deleg->ls_stateid), new_deleg, ls_hash);
2926			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2927			new_deleg = NULL;
2928			newnfsstats.srvdelegates++;
2929			nfsrv_openpluslock++;
2930			nfsrv_delegatecnt++;
2931		    }
2932		} else {
2933		    new_open->ls_stateid.seqid = 1;
2934		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2935		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2936		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2937		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
2938			NFSLCK_OPEN;
2939		    new_open->ls_uid = new_stp->ls_uid;
2940		    new_open->ls_openowner = ownerstp;
2941		    new_open->ls_lfp = lfp;
2942		    new_open->ls_clp = clp;
2943		    LIST_INIT(&new_open->ls_open);
2944		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2945		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2946		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2947			new_open, ls_hash);
2948		    openstp = new_open;
2949		    new_open = NULL;
2950		    newnfsstats.srvopens++;
2951		    nfsrv_openpluslock++;
2952
2953		    /*
2954		     * This is where we can choose to issue a delegation.
2955		     */
2956		    if (delegate == 0 || (writedeleg == 0 && readonly == 0) ||
2957			!NFSVNO_DELEGOK(vp) ||
2958			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2959			 LCL_CALLBACKSON)
2960			*rflagsp |= NFSV4OPEN_WDCONTENTION;
2961		    else if (nfsrv_issuedelegs == 0 ||
2962			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
2963			*rflagsp |= NFSV4OPEN_WDRESOURCE;
2964		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
2965			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
2966		    else {
2967			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2968			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2969			    = clp->lc_clientid.lval[0];
2970			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2971			    = clp->lc_clientid.lval[1];
2972			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2973			    = nfsrv_nextstateindex(clp);
2974			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
2975			    (nfsrv_writedelegifpos || !readonly) &&
2976			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
2977			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2978				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2979			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2980			} else {
2981			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2982				NFSLCK_READACCESS);
2983			    *rflagsp |= NFSV4OPEN_READDELEGATE;
2984			}
2985			new_deleg->ls_uid = new_stp->ls_uid;
2986			new_deleg->ls_lfp = lfp;
2987			new_deleg->ls_clp = clp;
2988			new_deleg->ls_filerev = filerev;
2989			new_deleg->ls_compref = nd->nd_compref;
2990			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2991			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2992			    new_deleg->ls_stateid), new_deleg, ls_hash);
2993			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2994			new_deleg = NULL;
2995			newnfsstats.srvdelegates++;
2996			nfsrv_openpluslock++;
2997			nfsrv_delegatecnt++;
2998		    }
2999		}
3000	} else {
3001		/*
3002		 * New owner case. Start the open_owner sequence with a
3003		 * Needs confirmation (unless a reclaim) and hang the
3004		 * new open off it.
3005		 */
3006		new_open->ls_stateid.seqid = 1;
3007		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3008		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3009		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3010		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3011		    NFSLCK_OPEN;
3012		new_open->ls_uid = new_stp->ls_uid;
3013		LIST_INIT(&new_open->ls_open);
3014		new_open->ls_openowner = new_stp;
3015		new_open->ls_lfp = lfp;
3016		new_open->ls_clp = clp;
3017		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3018		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3019			new_stp->ls_flags = 0;
3020		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3021			/* NFSv4.1 never needs confirmation. */
3022			new_stp->ls_flags = 0;
3023
3024			/*
3025			 * This is where we can choose to issue a delegation.
3026			 */
3027			if (delegate && nfsrv_issuedelegs &&
3028			    (writedeleg || readonly) &&
3029			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3030			     LCL_CALLBACKSON &&
3031			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3032			    NFSVNO_DELEGOK(vp) &&
3033			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3034			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3035				new_deleg->ls_stateid.seqid =
3036				    delegstateidp->seqid = 1;
3037				new_deleg->ls_stateid.other[0] =
3038				    delegstateidp->other[0]
3039				    = clp->lc_clientid.lval[0];
3040				new_deleg->ls_stateid.other[1] =
3041				    delegstateidp->other[1]
3042				    = clp->lc_clientid.lval[1];
3043				new_deleg->ls_stateid.other[2] =
3044				    delegstateidp->other[2]
3045				    = nfsrv_nextstateindex(clp);
3046				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3047				    (nfsrv_writedelegifpos || !readonly) &&
3048				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3049				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3050				     0)) {
3051					new_deleg->ls_flags =
3052					    (NFSLCK_DELEGWRITE |
3053					     NFSLCK_READACCESS |
3054					     NFSLCK_WRITEACCESS);
3055					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3056				} else {
3057					new_deleg->ls_flags =
3058					    (NFSLCK_DELEGREAD |
3059					     NFSLCK_READACCESS);
3060					*rflagsp |= NFSV4OPEN_READDELEGATE;
3061				}
3062				new_deleg->ls_uid = new_stp->ls_uid;
3063				new_deleg->ls_lfp = lfp;
3064				new_deleg->ls_clp = clp;
3065				new_deleg->ls_filerev = filerev;
3066				new_deleg->ls_compref = nd->nd_compref;
3067				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3068				    ls_file);
3069				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3070				    new_deleg->ls_stateid), new_deleg, ls_hash);
3071				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3072				    ls_list);
3073				new_deleg = NULL;
3074				newnfsstats.srvdelegates++;
3075				nfsrv_openpluslock++;
3076				nfsrv_delegatecnt++;
3077			}
3078		} else {
3079			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3080			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3081		}
3082		nfsrvd_refcache(new_stp->ls_op);
3083		new_stp->ls_noopens = 0;
3084		LIST_INIT(&new_stp->ls_open);
3085		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3086		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3087		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3088		    new_open, ls_hash);
3089		openstp = new_open;
3090		new_open = NULL;
3091		*new_stpp = NULL;
3092		newnfsstats.srvopens++;
3093		nfsrv_openpluslock++;
3094		newnfsstats.srvopenowners++;
3095		nfsrv_openpluslock++;
3096	}
3097	if (!error) {
3098		stateidp->seqid = openstp->ls_stateid.seqid;
3099		stateidp->other[0] = openstp->ls_stateid.other[0];
3100		stateidp->other[1] = openstp->ls_stateid.other[1];
3101		stateidp->other[2] = openstp->ls_stateid.other[2];
3102	}
3103	NFSUNLOCKSTATE();
3104	if (haslock) {
3105		NFSLOCKV4ROOTMUTEX();
3106		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3107		NFSUNLOCKV4ROOTMUTEX();
3108	}
3109	if (new_open)
3110		FREE((caddr_t)new_open, M_NFSDSTATE);
3111	if (new_deleg)
3112		FREE((caddr_t)new_deleg, M_NFSDSTATE);
3113
3114out:
3115	NFSEXITCODE2(error, nd);
3116	return (error);
3117}
3118
3119/*
3120 * Open update. Does the confirm, downgrade and close.
3121 */
3122APPLESTATIC int
3123nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3124    nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
3125{
3126	struct nfsstate *stp, *ownerstp;
3127	struct nfsclient *clp;
3128	struct nfslockfile *lfp;
3129	u_int32_t bits;
3130	int error = 0, gotstate = 0, len = 0;
3131	u_char client[NFSV4_OPAQUELIMIT];
3132
3133	/*
3134	 * Check for restart conditions (client and server).
3135	 */
3136	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3137	    &new_stp->ls_stateid, 0);
3138	if (error)
3139		goto out;
3140
3141	NFSLOCKSTATE();
3142	/*
3143	 * Get the open structure via clientid and stateid.
3144	 */
3145	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3146	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3147	if (!error)
3148		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3149		    new_stp->ls_flags, &stp);
3150
3151	/*
3152	 * Sanity check the open.
3153	 */
3154	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3155		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3156		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3157		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3158		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3159		error = NFSERR_BADSTATEID;
3160
3161	if (!error)
3162		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3163		    stp->ls_openowner, new_stp->ls_op);
3164	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3165	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3166	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3167	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3168	      new_stp->ls_stateid.seqid != 0)))
3169		error = NFSERR_OLDSTATEID;
3170	if (!error && vnode_vtype(vp) != VREG) {
3171		if (vnode_vtype(vp) == VDIR)
3172			error = NFSERR_ISDIR;
3173		else
3174			error = NFSERR_INVAL;
3175	}
3176
3177	if (error) {
3178		/*
3179		 * If a client tries to confirm an Open with a bad
3180		 * seqid# and there are no byte range locks or other Opens
3181		 * on the openowner, just throw it away, so the next use of the
3182		 * openowner will start a fresh seq#.
3183		 */
3184		if (error == NFSERR_BADSEQID &&
3185		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3186		    nfsrv_nootherstate(stp))
3187			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3188		NFSUNLOCKSTATE();
3189		goto out;
3190	}
3191
3192	/*
3193	 * Set the return stateid.
3194	 */
3195	stateidp->seqid = stp->ls_stateid.seqid + 1;
3196	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3197		stateidp->seqid = 1;
3198	stateidp->other[0] = stp->ls_stateid.other[0];
3199	stateidp->other[1] = stp->ls_stateid.other[1];
3200	stateidp->other[2] = stp->ls_stateid.other[2];
3201	/*
3202	 * Now, handle the three cases.
3203	 */
3204	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3205		/*
3206		 * If the open doesn't need confirmation, it seems to me that
3207		 * there is a client error, but I'll just log it and keep going?
3208		 */
3209		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3210			printf("Nfsv4d: stray open confirm\n");
3211		stp->ls_openowner->ls_flags = 0;
3212		stp->ls_stateid.seqid++;
3213		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3214		    stp->ls_stateid.seqid == 0)
3215			stp->ls_stateid.seqid = 1;
3216		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3217			clp->lc_flags |= LCL_STAMPEDSTABLE;
3218			len = clp->lc_idlen;
3219			NFSBCOPY(clp->lc_id, client, len);
3220			gotstate = 1;
3221		}
3222		NFSUNLOCKSTATE();
3223	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3224		ownerstp = stp->ls_openowner;
3225		lfp = stp->ls_lfp;
3226		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3227			/* Get the lf lock */
3228			nfsrv_locklf(lfp);
3229			NFSUNLOCKSTATE();
3230			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3231				NFSLOCKSTATE();
3232				nfsrv_unlocklf(lfp);
3233				NFSUNLOCKSTATE();
3234			}
3235		} else {
3236			(void) nfsrv_freeopen(stp, NULL, 0, p);
3237			NFSUNLOCKSTATE();
3238		}
3239	} else {
3240		/*
3241		 * Update the share bits, making sure that the new set are a
3242		 * subset of the old ones.
3243		 */
3244		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3245		if (~(stp->ls_flags) & bits) {
3246			NFSUNLOCKSTATE();
3247			error = NFSERR_INVAL;
3248			goto out;
3249		}
3250		stp->ls_flags = (bits | NFSLCK_OPEN);
3251		stp->ls_stateid.seqid++;
3252		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3253		    stp->ls_stateid.seqid == 0)
3254			stp->ls_stateid.seqid = 1;
3255		NFSUNLOCKSTATE();
3256	}
3257
3258	/*
3259	 * If the client just confirmed its first open, write a timestamp
3260	 * to the stable storage file.
3261	 */
3262	if (gotstate != 0) {
3263		nfsrv_writestable(client, len, NFSNST_NEWSTATE, p);
3264		nfsrv_backupstable();
3265	}
3266
3267out:
3268	NFSEXITCODE2(error, nd);
3269	return (error);
3270}
3271
3272/*
3273 * Delegation update. Does the purge and return.
3274 */
3275APPLESTATIC int
3276nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3277    nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3278    NFSPROC_T *p)
3279{
3280	struct nfsstate *stp;
3281	struct nfsclient *clp;
3282	int error = 0;
3283	fhandle_t fh;
3284
3285	/*
3286	 * Do a sanity check against the file handle for DelegReturn.
3287	 */
3288	if (vp) {
3289		error = nfsvno_getfh(vp, &fh, p);
3290		if (error)
3291			goto out;
3292	}
3293	/*
3294	 * Check for restart conditions (client and server).
3295	 */
3296	if (op == NFSV4OP_DELEGRETURN)
3297		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3298			stateidp, 0);
3299	else
3300		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3301			stateidp, 0);
3302
3303	NFSLOCKSTATE();
3304	/*
3305	 * Get the open structure via clientid and stateid.
3306	 */
3307	if (!error)
3308	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3309		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3310	if (error) {
3311		if (error == NFSERR_CBPATHDOWN)
3312			error = 0;
3313		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3314			error = NFSERR_STALESTATEID;
3315	}
3316	if (!error && op == NFSV4OP_DELEGRETURN) {
3317	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3318	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3319		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3320		error = NFSERR_OLDSTATEID;
3321	}
3322	/*
3323	 * NFSERR_EXPIRED means that the state has gone away,
3324	 * so Delegations have been purged. Just return ok.
3325	 */
3326	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3327		NFSUNLOCKSTATE();
3328		error = 0;
3329		goto out;
3330	}
3331	if (error) {
3332		NFSUNLOCKSTATE();
3333		goto out;
3334	}
3335
3336	if (op == NFSV4OP_DELEGRETURN) {
3337		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3338		    sizeof (fhandle_t))) {
3339			NFSUNLOCKSTATE();
3340			error = NFSERR_BADSTATEID;
3341			goto out;
3342		}
3343		nfsrv_freedeleg(stp);
3344	} else {
3345		nfsrv_freedeleglist(&clp->lc_olddeleg);
3346	}
3347	NFSUNLOCKSTATE();
3348	error = 0;
3349
3350out:
3351	NFSEXITCODE(error);
3352	return (error);
3353}
3354
3355/*
3356 * Release lock owner.
3357 */
3358APPLESTATIC int
3359nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3360    NFSPROC_T *p)
3361{
3362	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3363	struct nfsclient *clp;
3364	int error = 0;
3365
3366	/*
3367	 * Check for restart conditions (client and server).
3368	 */
3369	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3370	    &new_stp->ls_stateid, 0);
3371	if (error)
3372		goto out;
3373
3374	NFSLOCKSTATE();
3375	/*
3376	 * Get the lock owner by name.
3377	 */
3378	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3379	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3380	if (error) {
3381		NFSUNLOCKSTATE();
3382		goto out;
3383	}
3384	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3385	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3386		stp = LIST_FIRST(&openstp->ls_open);
3387		while (stp != LIST_END(&openstp->ls_open)) {
3388		    nstp = LIST_NEXT(stp, ls_list);
3389		    /*
3390		     * If the owner matches, check for locks and
3391		     * then free or return an error.
3392		     */
3393		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3394			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3395			 stp->ls_ownerlen)){
3396			if (LIST_EMPTY(&stp->ls_lock)) {
3397			    nfsrv_freelockowner(stp, NULL, 0, p);
3398			} else {
3399			    NFSUNLOCKSTATE();
3400			    error = NFSERR_LOCKSHELD;
3401			    goto out;
3402			}
3403		    }
3404		    stp = nstp;
3405		}
3406	    }
3407	}
3408	NFSUNLOCKSTATE();
3409
3410out:
3411	NFSEXITCODE(error);
3412	return (error);
3413}
3414
3415/*
3416 * Get the file handle for a lock structure.
3417 */
3418static int
3419nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3420    fhandle_t *nfhp, NFSPROC_T *p)
3421{
3422	fhandle_t *fhp = NULL;
3423	int error;
3424
3425	/*
3426	 * For lock, use the new nfslock structure, otherwise just
3427	 * a fhandle_t on the stack.
3428	 */
3429	if (flags & NFSLCK_OPEN) {
3430		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3431		fhp = &new_lfp->lf_fh;
3432	} else if (nfhp) {
3433		fhp = nfhp;
3434	} else {
3435		panic("nfsrv_getlockfh");
3436	}
3437	error = nfsvno_getfh(vp, fhp, p);
3438	NFSEXITCODE(error);
3439	return (error);
3440}
3441
3442/*
3443 * Get an nfs lock structure. Allocate one, as required, and return a
3444 * pointer to it.
3445 * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3446 */
3447static int
3448nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3449    struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3450{
3451	struct nfslockfile *lfp;
3452	fhandle_t *fhp = NULL, *tfhp;
3453	struct nfslockhashhead *hp;
3454	struct nfslockfile *new_lfp = NULL;
3455
3456	/*
3457	 * For lock, use the new nfslock structure, otherwise just
3458	 * a fhandle_t on the stack.
3459	 */
3460	if (flags & NFSLCK_OPEN) {
3461		new_lfp = *new_lfpp;
3462		fhp = &new_lfp->lf_fh;
3463	} else if (nfhp) {
3464		fhp = nfhp;
3465	} else {
3466		panic("nfsrv_getlockfile");
3467	}
3468
3469	hp = NFSLOCKHASH(fhp);
3470	LIST_FOREACH(lfp, hp, lf_hash) {
3471		tfhp = &lfp->lf_fh;
3472		if (NFSVNO_CMPFH(fhp, tfhp)) {
3473			if (lockit)
3474				nfsrv_locklf(lfp);
3475			*lfpp = lfp;
3476			return (0);
3477		}
3478	}
3479	if (!(flags & NFSLCK_OPEN))
3480		return (-1);
3481
3482	/*
3483	 * No match, so chain the new one into the list.
3484	 */
3485	LIST_INIT(&new_lfp->lf_open);
3486	LIST_INIT(&new_lfp->lf_lock);
3487	LIST_INIT(&new_lfp->lf_deleg);
3488	LIST_INIT(&new_lfp->lf_locallock);
3489	LIST_INIT(&new_lfp->lf_rollback);
3490	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3491	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3492	new_lfp->lf_usecount = 0;
3493	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3494	*lfpp = new_lfp;
3495	*new_lfpp = NULL;
3496	return (0);
3497}
3498
3499/*
3500 * This function adds a nfslock lock structure to the list for the associated
3501 * nfsstate and nfslockfile structures. It will be inserted after the
3502 * entry pointed at by insert_lop.
3503 */
3504static void
3505nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3506    struct nfsstate *stp, struct nfslockfile *lfp)
3507{
3508	struct nfslock *lop, *nlop;
3509
3510	new_lop->lo_stp = stp;
3511	new_lop->lo_lfp = lfp;
3512
3513	if (stp != NULL) {
3514		/* Insert in increasing lo_first order */
3515		lop = LIST_FIRST(&lfp->lf_lock);
3516		if (lop == LIST_END(&lfp->lf_lock) ||
3517		    new_lop->lo_first <= lop->lo_first) {
3518			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3519		} else {
3520			nlop = LIST_NEXT(lop, lo_lckfile);
3521			while (nlop != LIST_END(&lfp->lf_lock) &&
3522			       nlop->lo_first < new_lop->lo_first) {
3523				lop = nlop;
3524				nlop = LIST_NEXT(lop, lo_lckfile);
3525			}
3526			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3527		}
3528	} else {
3529		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3530	}
3531
3532	/*
3533	 * Insert after insert_lop, which is overloaded as stp or lfp for
3534	 * an empty list.
3535	 */
3536	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3537		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3538	else if ((struct nfsstate *)insert_lop == stp)
3539		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3540	else
3541		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3542	if (stp != NULL) {
3543		newnfsstats.srvlocks++;
3544		nfsrv_openpluslock++;
3545	}
3546}
3547
3548/*
3549 * This function updates the locking for a lock owner and given file. It
3550 * maintains a list of lock ranges ordered on increasing file offset that
3551 * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3552 * It always adds new_lop to the list and sometimes uses the one pointed
3553 * at by other_lopp.
3554 */
3555static void
3556nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3557    struct nfslock **other_lopp, struct nfslockfile *lfp)
3558{
3559	struct nfslock *new_lop = *new_lopp;
3560	struct nfslock *lop, *tlop, *ilop;
3561	struct nfslock *other_lop = *other_lopp;
3562	int unlock = 0, myfile = 0;
3563	u_int64_t tmp;
3564
3565	/*
3566	 * Work down the list until the lock is merged.
3567	 */
3568	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3569		unlock = 1;
3570	if (stp != NULL) {
3571		ilop = (struct nfslock *)stp;
3572		lop = LIST_FIRST(&stp->ls_lock);
3573	} else {
3574		ilop = (struct nfslock *)lfp;
3575		lop = LIST_FIRST(&lfp->lf_locallock);
3576	}
3577	while (lop != NULL) {
3578	    /*
3579	     * Only check locks for this file that aren't before the start of
3580	     * new lock's range.
3581	     */
3582	    if (lop->lo_lfp == lfp) {
3583	      myfile = 1;
3584	      if (lop->lo_end >= new_lop->lo_first) {
3585		if (new_lop->lo_end < lop->lo_first) {
3586			/*
3587			 * If the new lock ends before the start of the
3588			 * current lock's range, no merge, just insert
3589			 * the new lock.
3590			 */
3591			break;
3592		}
3593		if (new_lop->lo_flags == lop->lo_flags ||
3594		    (new_lop->lo_first <= lop->lo_first &&
3595		     new_lop->lo_end >= lop->lo_end)) {
3596			/*
3597			 * This lock can be absorbed by the new lock/unlock.
3598			 * This happens when it covers the entire range
3599			 * of the old lock or is contiguous
3600			 * with the old lock and is of the same type or an
3601			 * unlock.
3602			 */
3603			if (lop->lo_first < new_lop->lo_first)
3604				new_lop->lo_first = lop->lo_first;
3605			if (lop->lo_end > new_lop->lo_end)
3606				new_lop->lo_end = lop->lo_end;
3607			tlop = lop;
3608			lop = LIST_NEXT(lop, lo_lckowner);
3609			nfsrv_freenfslock(tlop);
3610			continue;
3611		}
3612
3613		/*
3614		 * All these cases are for contiguous locks that are not the
3615		 * same type, so they can't be merged.
3616		 */
3617		if (new_lop->lo_first <= lop->lo_first) {
3618			/*
3619			 * This case is where the new lock overlaps with the
3620			 * first part of the old lock. Move the start of the
3621			 * old lock to just past the end of the new lock. The
3622			 * new lock will be inserted in front of the old, since
3623			 * ilop hasn't been updated. (We are done now.)
3624			 */
3625			lop->lo_first = new_lop->lo_end;
3626			break;
3627		}
3628		if (new_lop->lo_end >= lop->lo_end) {
3629			/*
3630			 * This case is where the new lock overlaps with the
3631			 * end of the old lock's range. Move the old lock's
3632			 * end to just before the new lock's first and insert
3633			 * the new lock after the old lock.
3634			 * Might not be done yet, since the new lock could
3635			 * overlap further locks with higher ranges.
3636			 */
3637			lop->lo_end = new_lop->lo_first;
3638			ilop = lop;
3639			lop = LIST_NEXT(lop, lo_lckowner);
3640			continue;
3641		}
3642		/*
3643		 * The final case is where the new lock's range is in the
3644		 * middle of the current lock's and splits the current lock
3645		 * up. Use *other_lopp to handle the second part of the
3646		 * split old lock range. (We are done now.)
3647		 * For unlock, we use new_lop as other_lop and tmp, since
3648		 * other_lop and new_lop are the same for this case.
3649		 * We noted the unlock case above, so we don't need
3650		 * new_lop->lo_flags any longer.
3651		 */
3652		tmp = new_lop->lo_first;
3653		if (other_lop == NULL) {
3654			if (!unlock)
3655				panic("nfsd srv update unlock");
3656			other_lop = new_lop;
3657			*new_lopp = NULL;
3658		}
3659		other_lop->lo_first = new_lop->lo_end;
3660		other_lop->lo_end = lop->lo_end;
3661		other_lop->lo_flags = lop->lo_flags;
3662		other_lop->lo_stp = stp;
3663		other_lop->lo_lfp = lfp;
3664		lop->lo_end = tmp;
3665		nfsrv_insertlock(other_lop, lop, stp, lfp);
3666		*other_lopp = NULL;
3667		ilop = lop;
3668		break;
3669	      }
3670	    }
3671	    ilop = lop;
3672	    lop = LIST_NEXT(lop, lo_lckowner);
3673	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3674		break;
3675	}
3676
3677	/*
3678	 * Insert the new lock in the list at the appropriate place.
3679	 */
3680	if (!unlock) {
3681		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3682		*new_lopp = NULL;
3683	}
3684}
3685
3686/*
3687 * This function handles sequencing of locks, etc.
3688 * It returns an error that indicates what the caller should do.
3689 */
3690static int
3691nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3692    struct nfsstate *stp, struct nfsrvcache *op)
3693{
3694	int error = 0;
3695
3696	if ((nd->nd_flag & ND_NFSV41) != 0)
3697		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
3698		goto out;
3699	if (op != nd->nd_rp)
3700		panic("nfsrvstate checkseqid");
3701	if (!(op->rc_flag & RC_INPROG))
3702		panic("nfsrvstate not inprog");
3703	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3704		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3705		panic("nfsrvstate op refcnt");
3706	}
3707	if ((stp->ls_seq + 1) == seqid) {
3708		if (stp->ls_op)
3709			nfsrvd_derefcache(stp->ls_op);
3710		stp->ls_op = op;
3711		nfsrvd_refcache(op);
3712		stp->ls_seq = seqid;
3713		goto out;
3714	} else if (stp->ls_seq == seqid && stp->ls_op &&
3715		op->rc_xid == stp->ls_op->rc_xid &&
3716		op->rc_refcnt == 0 &&
3717		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3718		op->rc_cksum == stp->ls_op->rc_cksum) {
3719		if (stp->ls_op->rc_flag & RC_INPROG) {
3720			error = NFSERR_DONTREPLY;
3721			goto out;
3722		}
3723		nd->nd_rp = stp->ls_op;
3724		nd->nd_rp->rc_flag |= RC_INPROG;
3725		nfsrvd_delcache(op);
3726		error = NFSERR_REPLYFROMCACHE;
3727		goto out;
3728	}
3729	error = NFSERR_BADSEQID;
3730
3731out:
3732	NFSEXITCODE2(error, nd);
3733	return (error);
3734}
3735
3736/*
3737 * Get the client ip address for callbacks. If the strings can't be parsed,
3738 * just set lc_program to 0 to indicate no callbacks are possible.
3739 * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3740 *  the address to the client's transport address. This won't be used
3741 *  for callbacks, but can be printed out by newnfsstats for info.)
3742 * Return error if the xdr can't be parsed, 0 otherwise.
3743 */
3744APPLESTATIC int
3745nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3746{
3747	u_int32_t *tl;
3748	u_char *cp, *cp2;
3749	int i, j;
3750	struct sockaddr_in *rad, *sad;
3751	u_char protocol[5], addr[24];
3752	int error = 0, cantparse = 0;
3753	union {
3754		u_long ival;
3755		u_char cval[4];
3756	} ip;
3757	union {
3758		u_short sval;
3759		u_char cval[2];
3760	} port;
3761
3762	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3763	rad->sin_family = AF_INET;
3764	rad->sin_len = sizeof (struct sockaddr_in);
3765	rad->sin_addr.s_addr = 0;
3766	rad->sin_port = 0;
3767	clp->lc_req.nr_client = NULL;
3768	clp->lc_req.nr_lock = 0;
3769	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3770	i = fxdr_unsigned(int, *tl);
3771	if (i >= 3 && i <= 4) {
3772		error = nfsrv_mtostr(nd, protocol, i);
3773		if (error)
3774			goto nfsmout;
3775		if (!strcmp(protocol, "tcp")) {
3776			clp->lc_flags |= LCL_TCPCALLBACK;
3777			clp->lc_req.nr_sotype = SOCK_STREAM;
3778			clp->lc_req.nr_soproto = IPPROTO_TCP;
3779		} else if (!strcmp(protocol, "udp")) {
3780			clp->lc_req.nr_sotype = SOCK_DGRAM;
3781			clp->lc_req.nr_soproto = IPPROTO_UDP;
3782		} else {
3783			cantparse = 1;
3784		}
3785	} else {
3786		cantparse = 1;
3787		if (i > 0) {
3788			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3789			if (error)
3790				goto nfsmout;
3791		}
3792	}
3793	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3794	i = fxdr_unsigned(int, *tl);
3795	if (i < 0) {
3796		error = NFSERR_BADXDR;
3797		goto nfsmout;
3798	} else if (i == 0) {
3799		cantparse = 1;
3800	} else if (!cantparse && i <= 23 && i >= 11) {
3801		error = nfsrv_mtostr(nd, addr, i);
3802		if (error)
3803			goto nfsmout;
3804
3805		/*
3806		 * Parse out the address fields. We expect 6 decimal numbers
3807		 * separated by '.'s.
3808		 */
3809		cp = addr;
3810		i = 0;
3811		while (*cp && i < 6) {
3812			cp2 = cp;
3813			while (*cp2 && *cp2 != '.')
3814				cp2++;
3815			if (*cp2)
3816				*cp2++ = '\0';
3817			else if (i != 5) {
3818				cantparse = 1;
3819				break;
3820			}
3821			j = nfsrv_getipnumber(cp);
3822			if (j >= 0) {
3823				if (i < 4)
3824					ip.cval[3 - i] = j;
3825				else
3826					port.cval[5 - i] = j;
3827			} else {
3828				cantparse = 1;
3829				break;
3830			}
3831			cp = cp2;
3832			i++;
3833		}
3834		if (!cantparse) {
3835			if (ip.ival != 0x0) {
3836				rad->sin_addr.s_addr = htonl(ip.ival);
3837				rad->sin_port = htons(port.sval);
3838			} else {
3839				cantparse = 1;
3840			}
3841		}
3842	} else {
3843		cantparse = 1;
3844		if (i > 0) {
3845			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3846			if (error)
3847				goto nfsmout;
3848		}
3849	}
3850	if (cantparse) {
3851		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3852		rad->sin_addr.s_addr = sad->sin_addr.s_addr;
3853		rad->sin_port = 0x0;
3854		clp->lc_program = 0;
3855	}
3856nfsmout:
3857	NFSEXITCODE2(error, nd);
3858	return (error);
3859}
3860
3861/*
3862 * Turn a string of up to three decimal digits into a number. Return -1 upon
3863 * error.
3864 */
3865static int
3866nfsrv_getipnumber(u_char *cp)
3867{
3868	int i = 0, j = 0;
3869
3870	while (*cp) {
3871		if (j > 2 || *cp < '0' || *cp > '9')
3872			return (-1);
3873		i *= 10;
3874		i += (*cp - '0');
3875		cp++;
3876		j++;
3877	}
3878	if (i < 256)
3879		return (i);
3880	return (-1);
3881}
3882
3883/*
3884 * This function checks for restart conditions.
3885 */
3886static int
3887nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
3888    nfsv4stateid_t *stateidp, int specialid)
3889{
3890	int ret = 0;
3891
3892	/*
3893	 * First check for a server restart. Open, LockT, ReleaseLockOwner
3894	 * and DelegPurge have a clientid, the rest a stateid.
3895	 */
3896	if (flags &
3897	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
3898		if (clientid.lval[0] != nfsrvboottime) {
3899			ret = NFSERR_STALECLIENTID;
3900			goto out;
3901		}
3902	} else if (stateidp->other[0] != nfsrvboottime &&
3903		specialid == 0) {
3904		ret = NFSERR_STALESTATEID;
3905		goto out;
3906	}
3907
3908	/*
3909	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
3910	 * not use a lock/open owner seqid#, so the check can be done now.
3911	 * (The others will be checked, as required, later.)
3912	 */
3913	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
3914		goto out;
3915
3916	NFSLOCKSTATE();
3917	ret = nfsrv_checkgrace(NULL, NULL, flags);
3918	NFSUNLOCKSTATE();
3919
3920out:
3921	NFSEXITCODE(ret);
3922	return (ret);
3923}
3924
3925/*
3926 * Check for grace.
3927 */
3928static int
3929nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
3930    u_int32_t flags)
3931{
3932	int error = 0;
3933
3934	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
3935		if (flags & NFSLCK_RECLAIM) {
3936			error = NFSERR_NOGRACE;
3937			goto out;
3938		}
3939	} else {
3940		if (!(flags & NFSLCK_RECLAIM)) {
3941			error = NFSERR_GRACE;
3942			goto out;
3943		}
3944		if (nd != NULL && clp != NULL &&
3945		    (nd->nd_flag & ND_NFSV41) != 0 &&
3946		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
3947			error = NFSERR_NOGRACE;
3948			goto out;
3949		}
3950
3951		/*
3952		 * If grace is almost over and we are still getting Reclaims,
3953		 * extend grace a bit.
3954		 */
3955		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
3956		    nfsrv_stablefirst.nsf_eograce)
3957			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
3958				NFSRV_LEASEDELTA;
3959	}
3960
3961out:
3962	NFSEXITCODE(error);
3963	return (error);
3964}
3965
3966/*
3967 * Do a server callback.
3968 */
3969static int
3970nfsrv_docallback(struct nfsclient *clp, int procnum,
3971    nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
3972    struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
3973{
3974	mbuf_t m;
3975	u_int32_t *tl;
3976	struct nfsrv_descript nfsd, *nd = &nfsd;
3977	struct ucred *cred;
3978	int error = 0;
3979	u_int32_t callback;
3980	struct nfsdsession *sep = NULL;
3981
3982	cred = newnfs_getcred();
3983	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
3984	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
3985		NFSUNLOCKSTATE();
3986		panic("docallb");
3987	}
3988	clp->lc_cbref++;
3989
3990	/*
3991	 * Fill the callback program# and version into the request
3992	 * structure for newnfs_connect() to use.
3993	 */
3994	clp->lc_req.nr_prog = clp->lc_program;
3995#ifdef notnow
3996	if ((clp->lc_flags & LCL_NFSV41) != 0)
3997		clp->lc_req.nr_vers = NFSV41_CBVERS;
3998	else
3999#endif
4000		clp->lc_req.nr_vers = NFSV4_CBVERS;
4001
4002	/*
4003	 * First, fill in some of the fields of nd and cr.
4004	 */
4005	nd->nd_flag = ND_NFSV4;
4006	if (clp->lc_flags & LCL_GSS)
4007		nd->nd_flag |= ND_KERBV;
4008	if ((clp->lc_flags & LCL_NFSV41) != 0)
4009		nd->nd_flag |= ND_NFSV41;
4010	nd->nd_repstat = 0;
4011	cred->cr_uid = clp->lc_uid;
4012	cred->cr_gid = clp->lc_gid;
4013	callback = clp->lc_callback;
4014	NFSUNLOCKSTATE();
4015	cred->cr_ngroups = 1;
4016
4017	/*
4018	 * Get the first mbuf for the request.
4019	 */
4020	MGET(m, M_WAITOK, MT_DATA);
4021	mbuf_setlen(m, 0);
4022	nd->nd_mreq = nd->nd_mb = m;
4023	nd->nd_bpos = NFSMTOD(m, caddr_t);
4024
4025	/*
4026	 * and build the callback request.
4027	 */
4028	if (procnum == NFSV4OP_CBGETATTR) {
4029		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4030		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4031		    "CB Getattr", &sep);
4032		if (error != 0) {
4033			mbuf_freem(nd->nd_mreq);
4034			goto errout;
4035		}
4036		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4037		(void)nfsrv_putattrbit(nd, attrbitp);
4038	} else if (procnum == NFSV4OP_CBRECALL) {
4039		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4040		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4041		    "CB Recall", &sep);
4042		if (error != 0) {
4043			mbuf_freem(nd->nd_mreq);
4044			goto errout;
4045		}
4046		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4047		*tl++ = txdr_unsigned(stateidp->seqid);
4048		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4049		    NFSX_STATEIDOTHER);
4050		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4051		if (trunc)
4052			*tl = newnfs_true;
4053		else
4054			*tl = newnfs_false;
4055		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4056	} else if (procnum == NFSV4PROC_CBNULL) {
4057		nd->nd_procnum = NFSV4PROC_CBNULL;
4058		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4059			error = nfsv4_getcbsession(clp, &sep);
4060			if (error != 0) {
4061				mbuf_freem(nd->nd_mreq);
4062				goto errout;
4063			}
4064		}
4065	} else {
4066		error = NFSERR_SERVERFAULT;
4067		mbuf_freem(nd->nd_mreq);
4068		goto errout;
4069	}
4070
4071	/*
4072	 * Call newnfs_connect(), as required, and then newnfs_request().
4073	 */
4074	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4075	if (clp->lc_req.nr_client == NULL) {
4076		if ((clp->lc_flags & LCL_NFSV41) != 0)
4077			error = ECONNREFUSED;
4078		else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4079			error = newnfs_connect(NULL, &clp->lc_req, cred,
4080			    NULL, 1);
4081		else
4082			error = newnfs_connect(NULL, &clp->lc_req, cred,
4083			    NULL, 3);
4084	}
4085	newnfs_sndunlock(&clp->lc_req.nr_lock);
4086	if (!error) {
4087		if ((nd->nd_flag & ND_NFSV41) != 0) {
4088			KASSERT(sep != NULL, ("sep NULL"));
4089			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4090			    NULL, NULL, cred, clp->lc_program,
4091			    clp->lc_req.nr_vers, NULL, 1, NULL,
4092			    &sep->sess_cbsess);
4093			nfsrv_freesession(sep, NULL);
4094		} else
4095			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4096			    NULL, NULL, cred, clp->lc_program,
4097			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4098	}
4099errout:
4100	NFSFREECRED(cred);
4101
4102	/*
4103	 * If error is set here, the Callback path isn't working
4104	 * properly, so twiddle the appropriate LCL_ flags.
4105	 * (nd_repstat != 0 indicates the Callback path is working,
4106	 *  but the callback failed on the client.)
4107	 */
4108	if (error) {
4109		/*
4110		 * Mark the callback pathway down, which disabled issuing
4111		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4112		 */
4113		NFSLOCKSTATE();
4114		clp->lc_flags |= LCL_CBDOWN;
4115		NFSUNLOCKSTATE();
4116	} else {
4117		/*
4118		 * Callback worked. If the callback path was down, disable
4119		 * callbacks, so no more delegations will be issued. (This
4120		 * is done on the assumption that the callback pathway is
4121		 * flakey.)
4122		 */
4123		NFSLOCKSTATE();
4124		if (clp->lc_flags & LCL_CBDOWN)
4125			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4126		NFSUNLOCKSTATE();
4127		if (nd->nd_repstat)
4128			error = nd->nd_repstat;
4129		else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4130			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4131			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4132			    p, NULL);
4133		mbuf_freem(nd->nd_mrep);
4134	}
4135	NFSLOCKSTATE();
4136	clp->lc_cbref--;
4137	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4138		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4139		wakeup(clp);
4140	}
4141	NFSUNLOCKSTATE();
4142
4143	NFSEXITCODE(error);
4144	return (error);
4145}
4146
4147/*
4148 * Set up the compound RPC for the callback.
4149 */
4150static int
4151nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4152    uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4153{
4154	uint32_t *tl;
4155	int error, len;
4156
4157	len = strlen(optag);
4158	(void)nfsm_strtom(nd, optag, len);
4159	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4160	if ((nd->nd_flag & ND_NFSV41) != 0) {
4161		*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4162		*tl++ = txdr_unsigned(callback);
4163		*tl++ = txdr_unsigned(2);
4164		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4165		error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4166		if (error != 0)
4167			return (error);
4168		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4169		*tl = txdr_unsigned(op);
4170	} else {
4171		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4172		*tl++ = txdr_unsigned(callback);
4173		*tl++ = txdr_unsigned(1);
4174		*tl = txdr_unsigned(op);
4175	}
4176	return (0);
4177}
4178
4179/*
4180 * Return the next index# for a clientid. Mostly just increment and return
4181 * the next one, but... if the 32bit unsigned does actually wrap around,
4182 * it should be rebooted.
4183 * At an average rate of one new client per second, it will wrap around in
4184 * approximately 136 years. (I think the server will have been shut
4185 * down or rebooted before then.)
4186 */
4187static u_int32_t
4188nfsrv_nextclientindex(void)
4189{
4190	static u_int32_t client_index = 0;
4191
4192	client_index++;
4193	if (client_index != 0)
4194		return (client_index);
4195
4196	printf("%s: out of clientids\n", __func__);
4197	return (client_index);
4198}
4199
4200/*
4201 * Return the next index# for a stateid. Mostly just increment and return
4202 * the next one, but... if the 32bit unsigned does actually wrap around
4203 * (will a BSD server stay up that long?), find
4204 * new start and end values.
4205 */
4206static u_int32_t
4207nfsrv_nextstateindex(struct nfsclient *clp)
4208{
4209	struct nfsstate *stp;
4210	int i;
4211	u_int32_t canuse, min_index, max_index;
4212
4213	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4214		clp->lc_stateindex++;
4215		if (clp->lc_stateindex != clp->lc_statemaxindex)
4216			return (clp->lc_stateindex);
4217	}
4218
4219	/*
4220	 * Yuck, we've hit the end.
4221	 * Look for a new min and max.
4222	 */
4223	min_index = 0;
4224	max_index = 0xffffffff;
4225	for (i = 0; i < NFSSTATEHASHSIZE; i++) {
4226	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4227		if (stp->ls_stateid.other[2] > 0x80000000) {
4228		    if (stp->ls_stateid.other[2] < max_index)
4229			max_index = stp->ls_stateid.other[2];
4230		} else {
4231		    if (stp->ls_stateid.other[2] > min_index)
4232			min_index = stp->ls_stateid.other[2];
4233		}
4234	    }
4235	}
4236
4237	/*
4238	 * Yikes, highly unlikely, but I'll handle it anyhow.
4239	 */
4240	if (min_index == 0x80000000 && max_index == 0x80000001) {
4241	    canuse = 0;
4242	    /*
4243	     * Loop around until we find an unused entry. Return that
4244	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4245	     * (This is one of those rare cases where a goto is the
4246	     *  cleanest way to code the loop.)
4247	     */
4248tryagain:
4249	    for (i = 0; i < NFSSTATEHASHSIZE; i++) {
4250		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4251		    if (stp->ls_stateid.other[2] == canuse) {
4252			canuse++;
4253			goto tryagain;
4254		    }
4255		}
4256	    }
4257	    clp->lc_flags |= LCL_INDEXNOTOK;
4258	    return (canuse);
4259	}
4260
4261	/*
4262	 * Ok to start again from min + 1.
4263	 */
4264	clp->lc_stateindex = min_index + 1;
4265	clp->lc_statemaxindex = max_index;
4266	clp->lc_flags &= ~LCL_INDEXNOTOK;
4267	return (clp->lc_stateindex);
4268}
4269
4270/*
4271 * The following functions handle the stable storage file that deals with
4272 * the edge conditions described in RFC3530 Sec. 8.6.3.
4273 * The file is as follows:
4274 * - a single record at the beginning that has the lease time of the
4275 *   previous server instance (before the last reboot) and the nfsrvboottime
4276 *   values for the previous server boots.
4277 *   These previous boot times are used to ensure that the current
4278 *   nfsrvboottime does not, somehow, get set to a previous one.
4279 *   (This is important so that Stale ClientIDs and StateIDs can
4280 *    be recognized.)
4281 *   The number of previous nfsvrboottime values preceeds the list.
4282 * - followed by some number of appended records with:
4283 *   - client id string
4284 *   - flag that indicates it is a record revoking state via lease
4285 *     expiration or similar
4286 *     OR has successfully acquired state.
4287 * These structures vary in length, with the client string at the end, up
4288 * to NFSV4_OPAQUELIMIT in size.
4289 *
4290 * At the end of the grace period, the file is truncated, the first
4291 * record is rewritten with updated information and any acquired state
4292 * records for successful reclaims of state are written.
4293 *
4294 * Subsequent records are appended when the first state is issued to
4295 * a client and when state is revoked for a client.
4296 *
4297 * When reading the file in, state issued records that come later in
4298 * the file override older ones, since the append log is in cronological order.
4299 * If, for some reason, the file can't be read, the grace period is
4300 * immediately terminated and all reclaims get NFSERR_NOGRACE.
4301 */
4302
4303/*
4304 * Read in the stable storage file. Called by nfssvc() before the nfsd
4305 * processes start servicing requests.
4306 */
4307APPLESTATIC void
4308nfsrv_setupstable(NFSPROC_T *p)
4309{
4310	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4311	struct nfsrv_stable *sp, *nsp;
4312	struct nfst_rec *tsp;
4313	int error, i, tryagain;
4314	off_t off = 0;
4315	ssize_t aresid, len;
4316
4317	/*
4318	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4319	 * a reboot, so state has not been lost.
4320	 */
4321	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4322		return;
4323	/*
4324	 * Set Grace over just until the file reads successfully.
4325	 */
4326	nfsrvboottime = time_second;
4327	LIST_INIT(&sf->nsf_head);
4328	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4329	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4330	if (sf->nsf_fp == NULL)
4331		return;
4332	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4333	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4334	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4335	if (error || aresid || sf->nsf_numboots == 0 ||
4336		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4337		return;
4338
4339	/*
4340	 * Now, read in the boottimes.
4341	 */
4342	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4343		sizeof (time_t), M_TEMP, M_WAITOK);
4344	off = sizeof (struct nfsf_rec);
4345	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4346	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4347	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4348	if (error || aresid) {
4349		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4350		sf->nsf_bootvals = NULL;
4351		return;
4352	}
4353
4354	/*
4355	 * Make sure this nfsrvboottime is different from all recorded
4356	 * previous ones.
4357	 */
4358	do {
4359		tryagain = 0;
4360		for (i = 0; i < sf->nsf_numboots; i++) {
4361			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4362				nfsrvboottime++;
4363				tryagain = 1;
4364				break;
4365			}
4366		}
4367	} while (tryagain);
4368
4369	sf->nsf_flags |= NFSNSF_OK;
4370	off += (sf->nsf_numboots * sizeof (time_t));
4371
4372	/*
4373	 * Read through the file, building a list of records for grace
4374	 * checking.
4375	 * Each record is between sizeof (struct nfst_rec) and
4376	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4377	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4378	 */
4379	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4380		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4381	do {
4382	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4383	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4384	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4385	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4386	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4387		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4388		/*
4389		 * Yuck, the file has been corrupted, so just return
4390		 * after clearing out any restart state, so the grace period
4391		 * is over.
4392		 */
4393		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4394			LIST_REMOVE(sp, nst_list);
4395			free((caddr_t)sp, M_TEMP);
4396		}
4397		free((caddr_t)tsp, M_TEMP);
4398		sf->nsf_flags &= ~NFSNSF_OK;
4399		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4400		sf->nsf_bootvals = NULL;
4401		return;
4402	    }
4403	    if (len > 0) {
4404		off += sizeof (struct nfst_rec) + tsp->len - 1;
4405		/*
4406		 * Search the list for a matching client.
4407		 */
4408		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4409			if (tsp->len == sp->nst_len &&
4410			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4411				break;
4412		}
4413		if (sp == LIST_END(&sf->nsf_head)) {
4414			sp = (struct nfsrv_stable *)malloc(tsp->len +
4415				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4416				M_WAITOK);
4417			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4418				sizeof (struct nfst_rec) + tsp->len - 1);
4419			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4420		} else {
4421			if (tsp->flag == NFSNST_REVOKE)
4422				sp->nst_flag |= NFSNST_REVOKE;
4423			else
4424				/*
4425				 * A subsequent timestamp indicates the client
4426				 * did a setclientid/confirm and any previous
4427				 * revoke is no longer relevant.
4428				 */
4429				sp->nst_flag &= ~NFSNST_REVOKE;
4430		}
4431	    }
4432	} while (len > 0);
4433	free((caddr_t)tsp, M_TEMP);
4434	sf->nsf_flags = NFSNSF_OK;
4435	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4436		NFSRV_LEASEDELTA;
4437}
4438
4439/*
4440 * Update the stable storage file, now that the grace period is over.
4441 */
4442APPLESTATIC void
4443nfsrv_updatestable(NFSPROC_T *p)
4444{
4445	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4446	struct nfsrv_stable *sp, *nsp;
4447	int i;
4448	struct nfsvattr nva;
4449	vnode_t vp;
4450#if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4451	mount_t mp = NULL;
4452#endif
4453	int error;
4454
4455	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4456		return;
4457	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4458	/*
4459	 * Ok, we need to rewrite the stable storage file.
4460	 * - truncate to 0 length
4461	 * - write the new first structure
4462	 * - loop through the data structures, writing out any that
4463	 *   have timestamps older than the old boot
4464	 */
4465	if (sf->nsf_bootvals) {
4466		sf->nsf_numboots++;
4467		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4468			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4469	} else {
4470		sf->nsf_numboots = 1;
4471		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4472			M_TEMP, M_WAITOK);
4473	}
4474	sf->nsf_bootvals[0] = nfsrvboottime;
4475	sf->nsf_lease = nfsrv_lease;
4476	NFSVNO_ATTRINIT(&nva);
4477	NFSVNO_SETATTRVAL(&nva, size, 0);
4478	vp = NFSFPVNODE(sf->nsf_fp);
4479	vn_start_write(vp, &mp, V_WAIT);
4480	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4481		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4482		    NULL);
4483		NFSVOPUNLOCK(vp, 0);
4484	} else
4485		error = EPERM;
4486	vn_finished_write(mp);
4487	if (!error)
4488	    error = NFSD_RDWR(UIO_WRITE, vp,
4489		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4490		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4491	if (!error)
4492	    error = NFSD_RDWR(UIO_WRITE, vp,
4493		(caddr_t)sf->nsf_bootvals,
4494		sf->nsf_numboots * sizeof (time_t),
4495		(off_t)(sizeof (struct nfsf_rec)),
4496		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4497	free((caddr_t)sf->nsf_bootvals, M_TEMP);
4498	sf->nsf_bootvals = NULL;
4499	if (error) {
4500		sf->nsf_flags &= ~NFSNSF_OK;
4501		printf("EEK! Can't write NfsV4 stable storage file\n");
4502		return;
4503	}
4504	sf->nsf_flags |= NFSNSF_OK;
4505
4506	/*
4507	 * Loop through the list and write out timestamp records for
4508	 * any clients that successfully reclaimed state.
4509	 */
4510	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4511		if (sp->nst_flag & NFSNST_GOTSTATE) {
4512			nfsrv_writestable(sp->nst_client, sp->nst_len,
4513				NFSNST_NEWSTATE, p);
4514			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4515		}
4516		LIST_REMOVE(sp, nst_list);
4517		free((caddr_t)sp, M_TEMP);
4518	}
4519	nfsrv_backupstable();
4520}
4521
4522/*
4523 * Append a record to the stable storage file.
4524 */
4525APPLESTATIC void
4526nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4527{
4528	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4529	struct nfst_rec *sp;
4530	int error;
4531
4532	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4533		return;
4534	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4535		len - 1, M_TEMP, M_WAITOK);
4536	sp->len = len;
4537	NFSBCOPY(client, sp->client, len);
4538	sp->flag = flag;
4539	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4540	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4541	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4542	free((caddr_t)sp, M_TEMP);
4543	if (error) {
4544		sf->nsf_flags &= ~NFSNSF_OK;
4545		printf("EEK! Can't write NfsV4 stable storage file\n");
4546	}
4547}
4548
4549/*
4550 * This function is called during the grace period to mark a client
4551 * that successfully reclaimed state.
4552 */
4553static void
4554nfsrv_markstable(struct nfsclient *clp)
4555{
4556	struct nfsrv_stable *sp;
4557
4558	/*
4559	 * First find the client structure.
4560	 */
4561	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4562		if (sp->nst_len == clp->lc_idlen &&
4563		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4564			break;
4565	}
4566	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4567		return;
4568
4569	/*
4570	 * Now, just mark it and set the nfsclient back pointer.
4571	 */
4572	sp->nst_flag |= NFSNST_GOTSTATE;
4573	sp->nst_clp = clp;
4574}
4575
4576/*
4577 * This function is called for a reclaim, to see if it gets grace.
4578 * It returns 0 if a reclaim is allowed, 1 otherwise.
4579 */
4580static int
4581nfsrv_checkstable(struct nfsclient *clp)
4582{
4583	struct nfsrv_stable *sp;
4584
4585	/*
4586	 * First, find the entry for the client.
4587	 */
4588	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4589		if (sp->nst_len == clp->lc_idlen &&
4590		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4591			break;
4592	}
4593
4594	/*
4595	 * If not in the list, state was revoked or no state was issued
4596	 * since the previous reboot, a reclaim is denied.
4597	 */
4598	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4599	    (sp->nst_flag & NFSNST_REVOKE) ||
4600	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4601		return (1);
4602	return (0);
4603}
4604
4605/*
4606 * Test for and try to clear out a conflicting client. This is called by
4607 * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4608 * a found.
4609 * The trick here is that it can't revoke a conflicting client with an
4610 * expired lease unless it holds the v4root lock, so...
4611 * If no v4root lock, get the lock and return 1 to indicate "try again".
4612 * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4613 * the revocation worked and the conflicting client is "bye, bye", so it
4614 * can be tried again.
4615 * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4616 * Unlocks State before a non-zero value is returned.
4617 */
4618static int
4619nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4620    NFSPROC_T *p)
4621{
4622	int gotlock, lktype;
4623
4624	/*
4625	 * If lease hasn't expired, we can't fix it.
4626	 */
4627	if (clp->lc_expiry >= NFSD_MONOSEC ||
4628	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4629		return (0);
4630	if (*haslockp == 0) {
4631		NFSUNLOCKSTATE();
4632		lktype = NFSVOPISLOCKED(vp);
4633		NFSVOPUNLOCK(vp, 0);
4634		NFSLOCKV4ROOTMUTEX();
4635		nfsv4_relref(&nfsv4rootfs_lock);
4636		do {
4637			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4638			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4639		} while (!gotlock);
4640		NFSUNLOCKV4ROOTMUTEX();
4641		*haslockp = 1;
4642		NFSVOPLOCK(vp, lktype | LK_RETRY);
4643		if ((vp->v_iflag & VI_DOOMED) != 0)
4644			return (2);
4645		else
4646			return (1);
4647	}
4648	NFSUNLOCKSTATE();
4649
4650	/*
4651	 * Ok, we can expire the conflicting client.
4652	 */
4653	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4654	nfsrv_backupstable();
4655	nfsrv_cleanclient(clp, p);
4656	nfsrv_freedeleglist(&clp->lc_deleg);
4657	nfsrv_freedeleglist(&clp->lc_olddeleg);
4658	LIST_REMOVE(clp, lc_hash);
4659	nfsrv_zapclient(clp, p);
4660	return (1);
4661}
4662
4663/*
4664 * Resolve a delegation conflict.
4665 * Returns 0 to indicate the conflict was resolved without sleeping.
4666 * Return -1 to indicate that the caller should check for conflicts again.
4667 * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4668 *
4669 * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4670 * for a return of 0, since there was no sleep and it could be required
4671 * later. It is released for a return of NFSERR_DELAY, since the caller
4672 * will return that error. It is released when a sleep was done waiting
4673 * for the delegation to be returned or expire (so that other nfsds can
4674 * handle ops). Then, it must be acquired for the write to stable storage.
4675 * (This function is somewhat similar to nfsrv_clientconflict(), but
4676 *  the semantics differ in a couple of subtle ways. The return of 0
4677 *  indicates the conflict was resolved without sleeping here, not
4678 *  that the conflict can't be resolved and the handling of nfsv4root_lock
4679 *  differs, as noted above.)
4680 * Unlocks State before returning a non-zero value.
4681 */
4682static int
4683nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4684    vnode_t vp)
4685{
4686	struct nfsclient *clp = stp->ls_clp;
4687	int gotlock, error, lktype, retrycnt, zapped_clp;
4688	nfsv4stateid_t tstateid;
4689	fhandle_t tfh;
4690
4691	/*
4692	 * If the conflict is with an old delegation...
4693	 */
4694	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4695		/*
4696		 * You can delete it, if it has expired.
4697		 */
4698		if (clp->lc_delegtime < NFSD_MONOSEC) {
4699			nfsrv_freedeleg(stp);
4700			NFSUNLOCKSTATE();
4701			error = -1;
4702			goto out;
4703		}
4704		NFSUNLOCKSTATE();
4705		/*
4706		 * During this delay, the old delegation could expire or it
4707		 * could be recovered by the client via an Open with
4708		 * CLAIM_DELEGATE_PREV.
4709		 * Release the nfsv4root_lock, if held.
4710		 */
4711		if (*haslockp) {
4712			*haslockp = 0;
4713			NFSLOCKV4ROOTMUTEX();
4714			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4715			NFSUNLOCKV4ROOTMUTEX();
4716		}
4717		error = NFSERR_DELAY;
4718		goto out;
4719	}
4720
4721	/*
4722	 * It's a current delegation, so:
4723	 * - check to see if the delegation has expired
4724	 *   - if so, get the v4root lock and then expire it
4725	 */
4726	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4727		/*
4728		 * - do a recall callback, since not yet done
4729		 * For now, never allow truncate to be set. To use
4730		 * truncate safely, it must be guaranteed that the
4731		 * Remove, Rename or Setattr with size of 0 will
4732		 * succeed and that would require major changes to
4733		 * the VFS/Vnode OPs.
4734		 * Set the expiry time large enough so that it won't expire
4735		 * until after the callback, then set it correctly, once
4736		 * the callback is done. (The delegation will now time
4737		 * out whether or not the Recall worked ok. The timeout
4738		 * will be extended when ops are done on the delegation
4739		 * stateid, up to the timelimit.)
4740		 */
4741		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4742		    NFSRV_LEASEDELTA;
4743		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4744		    NFSRV_LEASEDELTA;
4745		stp->ls_flags |= NFSLCK_DELEGRECALL;
4746
4747		/*
4748		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4749		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4750		 * in order to try and avoid a race that could happen
4751		 * when a CBRecall request passed the Open reply with
4752		 * the delegation in it when transitting the network.
4753		 * Since nfsrv_docallback will sleep, don't use stp after
4754		 * the call.
4755		 */
4756		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4757		    sizeof (tstateid));
4758		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4759		    sizeof (tfh));
4760		NFSUNLOCKSTATE();
4761		if (*haslockp) {
4762			*haslockp = 0;
4763			NFSLOCKV4ROOTMUTEX();
4764			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4765			NFSUNLOCKV4ROOTMUTEX();
4766		}
4767		retrycnt = 0;
4768		do {
4769		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4770			&tstateid, 0, &tfh, NULL, NULL, p);
4771		    retrycnt++;
4772		} while ((error == NFSERR_BADSTATEID ||
4773		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4774		error = NFSERR_DELAY;
4775		goto out;
4776	}
4777
4778	if (clp->lc_expiry >= NFSD_MONOSEC &&
4779	    stp->ls_delegtime >= NFSD_MONOSEC) {
4780		NFSUNLOCKSTATE();
4781		/*
4782		 * A recall has been done, but it has not yet expired.
4783		 * So, RETURN_DELAY.
4784		 */
4785		if (*haslockp) {
4786			*haslockp = 0;
4787			NFSLOCKV4ROOTMUTEX();
4788			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4789			NFSUNLOCKV4ROOTMUTEX();
4790		}
4791		error = NFSERR_DELAY;
4792		goto out;
4793	}
4794
4795	/*
4796	 * If we don't yet have the lock, just get it and then return,
4797	 * since we need that before deleting expired state, such as
4798	 * this delegation.
4799	 * When getting the lock, unlock the vnode, so other nfsds that
4800	 * are in progress, won't get stuck waiting for the vnode lock.
4801	 */
4802	if (*haslockp == 0) {
4803		NFSUNLOCKSTATE();
4804		lktype = NFSVOPISLOCKED(vp);
4805		NFSVOPUNLOCK(vp, 0);
4806		NFSLOCKV4ROOTMUTEX();
4807		nfsv4_relref(&nfsv4rootfs_lock);
4808		do {
4809			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4810			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4811		} while (!gotlock);
4812		NFSUNLOCKV4ROOTMUTEX();
4813		*haslockp = 1;
4814		NFSVOPLOCK(vp, lktype | LK_RETRY);
4815		if ((vp->v_iflag & VI_DOOMED) != 0) {
4816			*haslockp = 0;
4817			NFSLOCKV4ROOTMUTEX();
4818			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4819			NFSUNLOCKV4ROOTMUTEX();
4820			error = NFSERR_PERM;
4821			goto out;
4822		}
4823		error = -1;
4824		goto out;
4825	}
4826
4827	NFSUNLOCKSTATE();
4828	/*
4829	 * Ok, we can delete the expired delegation.
4830	 * First, write the Revoke record to stable storage and then
4831	 * clear out the conflict.
4832	 * Since all other nfsd threads are now blocked, we can safely
4833	 * sleep without the state changing.
4834	 */
4835	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4836	nfsrv_backupstable();
4837	if (clp->lc_expiry < NFSD_MONOSEC) {
4838		nfsrv_cleanclient(clp, p);
4839		nfsrv_freedeleglist(&clp->lc_deleg);
4840		nfsrv_freedeleglist(&clp->lc_olddeleg);
4841		LIST_REMOVE(clp, lc_hash);
4842		zapped_clp = 1;
4843	} else {
4844		nfsrv_freedeleg(stp);
4845		zapped_clp = 0;
4846	}
4847	if (zapped_clp)
4848		nfsrv_zapclient(clp, p);
4849	error = -1;
4850
4851out:
4852	NFSEXITCODE(error);
4853	return (error);
4854}
4855
4856/*
4857 * Check for a remove allowed, if remove is set to 1 and get rid of
4858 * delegations.
4859 */
4860APPLESTATIC int
4861nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
4862{
4863	struct nfsstate *stp;
4864	struct nfslockfile *lfp;
4865	int error, haslock = 0;
4866	fhandle_t nfh;
4867
4868	/*
4869	 * First, get the lock file structure.
4870	 * (A return of -1 means no associated state, so remove ok.)
4871	 */
4872	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4873tryagain:
4874	NFSLOCKSTATE();
4875	if (!error)
4876		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
4877	if (error) {
4878		NFSUNLOCKSTATE();
4879		if (haslock) {
4880			NFSLOCKV4ROOTMUTEX();
4881			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4882			NFSUNLOCKV4ROOTMUTEX();
4883		}
4884		if (error == -1)
4885			error = 0;
4886		goto out;
4887	}
4888
4889	/*
4890	 * Now, we must Recall any delegations.
4891	 */
4892	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
4893	if (error) {
4894		/*
4895		 * nfsrv_cleandeleg() unlocks state for non-zero
4896		 * return.
4897		 */
4898		if (error == -1)
4899			goto tryagain;
4900		if (haslock) {
4901			NFSLOCKV4ROOTMUTEX();
4902			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4903			NFSUNLOCKV4ROOTMUTEX();
4904		}
4905		goto out;
4906	}
4907
4908	/*
4909	 * Now, look for a conflicting open share.
4910	 */
4911	if (remove) {
4912		LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
4913			if (stp->ls_flags & NFSLCK_WRITEDENY) {
4914				error = NFSERR_FILEOPEN;
4915				break;
4916			}
4917		}
4918	}
4919
4920	NFSUNLOCKSTATE();
4921	if (haslock) {
4922		NFSLOCKV4ROOTMUTEX();
4923		nfsv4_unlock(&nfsv4rootfs_lock, 1);
4924		NFSUNLOCKV4ROOTMUTEX();
4925	}
4926
4927out:
4928	NFSEXITCODE(error);
4929	return (error);
4930}
4931
4932/*
4933 * Clear out all delegations for the file referred to by lfp.
4934 * May return NFSERR_DELAY, if there will be a delay waiting for
4935 * delegations to expire.
4936 * Returns -1 to indicate it slept while recalling a delegation.
4937 * This function has the side effect of deleting the nfslockfile structure,
4938 * if it no longer has associated state and didn't have to sleep.
4939 * Unlocks State before a non-zero value is returned.
4940 */
4941static int
4942nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
4943    struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
4944{
4945	struct nfsstate *stp, *nstp;
4946	int ret = 0;
4947
4948	stp = LIST_FIRST(&lfp->lf_deleg);
4949	while (stp != LIST_END(&lfp->lf_deleg)) {
4950		nstp = LIST_NEXT(stp, ls_file);
4951		if (stp->ls_clp != clp) {
4952			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
4953			if (ret) {
4954				/*
4955				 * nfsrv_delegconflict() unlocks state
4956				 * when it returns non-zero.
4957				 */
4958				goto out;
4959			}
4960		}
4961		stp = nstp;
4962	}
4963out:
4964	NFSEXITCODE(ret);
4965	return (ret);
4966}
4967
4968/*
4969 * There are certain operations that, when being done outside of NFSv4,
4970 * require that any NFSv4 delegation for the file be recalled.
4971 * This function is to be called for those cases:
4972 * VOP_RENAME() - When a delegation is being recalled for any reason,
4973 *	the client may have to do Opens against the server, using the file's
4974 *	final component name. If the file has been renamed on the server,
4975 *	that component name will be incorrect and the Open will fail.
4976 * VOP_REMOVE() - Theoretically, a client could Open a file after it has
4977 *	been removed on the server, if there is a delegation issued to
4978 *	that client for the file. I say "theoretically" since clients
4979 *	normally do an Access Op before the Open and that Access Op will
4980 *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
4981 *	they will detect the file's removal in the same manner. (There is
4982 *	one case where RFC3530 allows a client to do an Open without first
4983 *	doing an Access Op, which is passage of a check against the ACE
4984 *	returned with a Write delegation, but current practice is to ignore
4985 *	the ACE and always do an Access Op.)
4986 *	Since the functions can only be called with an unlocked vnode, this
4987 *	can't be done at this time.
4988 * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
4989 *	locks locally in the client, which are not visible to the server. To
4990 *	deal with this, issuing of delegations for a vnode must be disabled
4991 *	and all delegations for the vnode recalled. This is done via the
4992 *	second function, using the VV_DISABLEDELEG vflag on the vnode.
4993 */
4994APPLESTATIC void
4995nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
4996{
4997	time_t starttime;
4998	int error;
4999
5000	/*
5001	 * First, check to see if the server is currently running and it has
5002	 * been called for a regular file when issuing delegations.
5003	 */
5004	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5005	    nfsrv_issuedelegs == 0)
5006		return;
5007
5008	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5009	/*
5010	 * First, get a reference on the nfsv4rootfs_lock so that an
5011	 * exclusive lock cannot be acquired by another thread.
5012	 */
5013	NFSLOCKV4ROOTMUTEX();
5014	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5015	NFSUNLOCKV4ROOTMUTEX();
5016
5017	/*
5018	 * Now, call nfsrv_checkremove() in a loop while it returns
5019	 * NFSERR_DELAY. Return upon any other error or when timed out.
5020	 */
5021	starttime = NFSD_MONOSEC;
5022	do {
5023		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5024			error = nfsrv_checkremove(vp, 0, p);
5025			NFSVOPUNLOCK(vp, 0);
5026		} else
5027			error = EPERM;
5028		if (error == NFSERR_DELAY) {
5029			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5030				break;
5031			/* Sleep for a short period of time */
5032			(void) nfs_catnap(PZERO, 0, "nfsremove");
5033		}
5034	} while (error == NFSERR_DELAY);
5035	NFSLOCKV4ROOTMUTEX();
5036	nfsv4_relref(&nfsv4rootfs_lock);
5037	NFSUNLOCKV4ROOTMUTEX();
5038}
5039
5040APPLESTATIC void
5041nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5042{
5043
5044#ifdef VV_DISABLEDELEG
5045	/*
5046	 * First, flag issuance of delegations disabled.
5047	 */
5048	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5049#endif
5050
5051	/*
5052	 * Then call nfsd_recalldelegation() to get rid of all extant
5053	 * delegations.
5054	 */
5055	nfsd_recalldelegation(vp, p);
5056}
5057
5058/*
5059 * Check for conflicting locks, etc. and then get rid of delegations.
5060 * (At one point I thought that I should get rid of delegations for any
5061 *  Setattr, since it could potentially disallow the I/O op (read or write)
5062 *  allowed by the delegation. However, Setattr Ops that aren't changing
5063 *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5064 *  for the same client or a different one, so I decided to only get rid
5065 *  of delegations for other clients when the size is being changed.)
5066 * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5067 * as Write backs, even if there is no delegation, so it really isn't any
5068 * different?)
5069 */
5070APPLESTATIC int
5071nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5072    nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5073    struct nfsexstuff *exp, NFSPROC_T *p)
5074{
5075	struct nfsstate st, *stp = &st;
5076	struct nfslock lo, *lop = &lo;
5077	int error = 0;
5078	nfsquad_t clientid;
5079
5080	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5081		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5082		lop->lo_first = nvap->na_size;
5083	} else {
5084		stp->ls_flags = 0;
5085		lop->lo_first = 0;
5086	}
5087	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5088	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5089	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5090	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5091		stp->ls_flags |= NFSLCK_SETATTR;
5092	if (stp->ls_flags == 0)
5093		goto out;
5094	lop->lo_end = NFS64BITSSET;
5095	lop->lo_flags = NFSLCK_WRITE;
5096	stp->ls_ownerlen = 0;
5097	stp->ls_op = NULL;
5098	stp->ls_uid = nd->nd_cred->cr_uid;
5099	stp->ls_stateid.seqid = stateidp->seqid;
5100	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5101	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5102	stp->ls_stateid.other[2] = stateidp->other[2];
5103	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5104	    stateidp, exp, nd, p);
5105
5106out:
5107	NFSEXITCODE2(error, nd);
5108	return (error);
5109}
5110
5111/*
5112 * Check for a write delegation and do a CBGETATTR if there is one, updating
5113 * the attributes, as required.
5114 * Should I return an error if I can't get the attributes? (For now, I'll
5115 * just return ok.
5116 */
5117APPLESTATIC int
5118nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5119    struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
5120    NFSPROC_T *p)
5121{
5122	struct nfsstate *stp;
5123	struct nfslockfile *lfp;
5124	struct nfsclient *clp;
5125	struct nfsvattr nva;
5126	fhandle_t nfh;
5127	int error = 0;
5128	nfsattrbit_t cbbits;
5129	u_quad_t delegfilerev;
5130
5131	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5132	if (!NFSNONZERO_ATTRBIT(&cbbits))
5133		goto out;
5134
5135	/*
5136	 * Get the lock file structure.
5137	 * (A return of -1 means no associated state, so return ok.)
5138	 */
5139	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5140	NFSLOCKSTATE();
5141	if (!error)
5142		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5143	if (error) {
5144		NFSUNLOCKSTATE();
5145		if (error == -1)
5146			error = 0;
5147		goto out;
5148	}
5149
5150	/*
5151	 * Now, look for a write delegation.
5152	 */
5153	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5154		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5155			break;
5156	}
5157	if (stp == LIST_END(&lfp->lf_deleg)) {
5158		NFSUNLOCKSTATE();
5159		goto out;
5160	}
5161	clp = stp->ls_clp;
5162	delegfilerev = stp->ls_filerev;
5163
5164	/*
5165	 * If the Write delegation was issued as a part of this Compound RPC
5166	 * or if we have an Implied Clientid (used in a previous Op in this
5167	 * compound) and it is the client the delegation was issued to,
5168	 * just return ok.
5169	 * I also assume that it is from the same client iff the network
5170	 * host IP address is the same as the callback address. (Not
5171	 * exactly correct by the RFC, but avoids a lot of Getattr
5172	 * callbacks.)
5173	 */
5174	if (nd->nd_compref == stp->ls_compref ||
5175	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5176	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5177	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5178		NFSUNLOCKSTATE();
5179		goto out;
5180	}
5181
5182	/*
5183	 * We are now done with the delegation state structure,
5184	 * so the statelock can be released and we can now tsleep().
5185	 */
5186
5187	/*
5188	 * Now, we must do the CB Getattr callback, to see if Change or Size
5189	 * has changed.
5190	 */
5191	if (clp->lc_expiry >= NFSD_MONOSEC) {
5192		NFSUNLOCKSTATE();
5193		NFSVNO_ATTRINIT(&nva);
5194		nva.na_filerev = NFS64BITSSET;
5195		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5196		    0, &nfh, &nva, &cbbits, p);
5197		if (!error) {
5198			if ((nva.na_filerev != NFS64BITSSET &&
5199			    nva.na_filerev > delegfilerev) ||
5200			    (NFSVNO_ISSETSIZE(&nva) &&
5201			     nva.na_size != nvap->na_size)) {
5202				error = nfsvno_updfilerev(vp, nvap, cred, p);
5203				if (NFSVNO_ISSETSIZE(&nva))
5204					nvap->na_size = nva.na_size;
5205			}
5206		} else
5207			error = 0;	/* Ignore callback errors for now. */
5208	} else {
5209		NFSUNLOCKSTATE();
5210	}
5211
5212out:
5213	NFSEXITCODE2(error, nd);
5214	return (error);
5215}
5216
5217/*
5218 * This function looks for openowners that haven't had any opens for
5219 * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5220 * is set.
5221 */
5222APPLESTATIC void
5223nfsrv_throwawayopens(NFSPROC_T *p)
5224{
5225	struct nfsclient *clp, *nclp;
5226	struct nfsstate *stp, *nstp;
5227	int i;
5228
5229	NFSLOCKSTATE();
5230	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5231	/*
5232	 * For each client...
5233	 */
5234	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
5235	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5236		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5237			if (LIST_EMPTY(&stp->ls_open) &&
5238			    (stp->ls_noopens > NFSNOOPEN ||
5239			     (nfsrv_openpluslock * 2) >
5240			     NFSRV_V4STATELIMIT))
5241				nfsrv_freeopenowner(stp, 0, p);
5242		}
5243	    }
5244	}
5245	NFSUNLOCKSTATE();
5246}
5247
5248/*
5249 * This function checks to see if the credentials are the same.
5250 * Returns 1 for not same, 0 otherwise.
5251 */
5252static int
5253nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5254{
5255
5256	if (nd->nd_flag & ND_GSS) {
5257		if (!(clp->lc_flags & LCL_GSS))
5258			return (1);
5259		if (clp->lc_flags & LCL_NAME) {
5260			if (nd->nd_princlen != clp->lc_namelen ||
5261			    NFSBCMP(nd->nd_principal, clp->lc_name,
5262				clp->lc_namelen))
5263				return (1);
5264			else
5265				return (0);
5266		}
5267		if (nd->nd_cred->cr_uid == clp->lc_uid)
5268			return (0);
5269		else
5270			return (1);
5271	} else if (clp->lc_flags & LCL_GSS)
5272		return (1);
5273	/*
5274	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5275	 * in RFC3530, which talks about principals, but doesn't say anything
5276	 * about uids for AUTH_SYS.)
5277	 */
5278	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5279		return (0);
5280	else
5281		return (1);
5282}
5283
5284/*
5285 * Calculate the lease expiry time.
5286 */
5287static time_t
5288nfsrv_leaseexpiry(void)
5289{
5290
5291	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5292		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5293	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5294}
5295
5296/*
5297 * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5298 */
5299static void
5300nfsrv_delaydelegtimeout(struct nfsstate *stp)
5301{
5302
5303	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5304		return;
5305
5306	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5307	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5308		stp->ls_delegtime += nfsrv_lease;
5309		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5310			stp->ls_delegtime = stp->ls_delegtimelimit;
5311	}
5312}
5313
5314/*
5315 * This function checks to see if there is any other state associated
5316 * with the openowner for this Open.
5317 * It returns 1 if there is no other state, 0 otherwise.
5318 */
5319static int
5320nfsrv_nootherstate(struct nfsstate *stp)
5321{
5322	struct nfsstate *tstp;
5323
5324	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5325		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5326			return (0);
5327	}
5328	return (1);
5329}
5330
5331/*
5332 * Create a list of lock deltas (changes to local byte range locking
5333 * that can be rolled back using the list) and apply the changes via
5334 * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5335 * the rollback or update function will be called after this.
5336 * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5337 * call fails. If it returns an error, it will unlock the list.
5338 */
5339static int
5340nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5341    uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5342{
5343	struct nfslock *lop, *nlop;
5344	int error = 0;
5345
5346	/* Loop through the list of locks. */
5347	lop = LIST_FIRST(&lfp->lf_locallock);
5348	while (first < end && lop != NULL) {
5349		nlop = LIST_NEXT(lop, lo_lckowner);
5350		if (first >= lop->lo_end) {
5351			/* not there yet */
5352			lop = nlop;
5353		} else if (first < lop->lo_first) {
5354			/* new one starts before entry in list */
5355			if (end <= lop->lo_first) {
5356				/* no overlap between old and new */
5357				error = nfsrv_dolocal(vp, lfp, flags,
5358				    NFSLCK_UNLOCK, first, end, cfp, p);
5359				if (error != 0)
5360					break;
5361				first = end;
5362			} else {
5363				/* handle fragment overlapped with new one */
5364				error = nfsrv_dolocal(vp, lfp, flags,
5365				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5366				    p);
5367				if (error != 0)
5368					break;
5369				first = lop->lo_first;
5370			}
5371		} else {
5372			/* new one overlaps this entry in list */
5373			if (end <= lop->lo_end) {
5374				/* overlaps all of new one */
5375				error = nfsrv_dolocal(vp, lfp, flags,
5376				    lop->lo_flags, first, end, cfp, p);
5377				if (error != 0)
5378					break;
5379				first = end;
5380			} else {
5381				/* handle fragment overlapped with new one */
5382				error = nfsrv_dolocal(vp, lfp, flags,
5383				    lop->lo_flags, first, lop->lo_end, cfp, p);
5384				if (error != 0)
5385					break;
5386				first = lop->lo_end;
5387				lop = nlop;
5388			}
5389		}
5390	}
5391	if (first < end && error == 0)
5392		/* handle fragment past end of list */
5393		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5394		    end, cfp, p);
5395
5396	NFSEXITCODE(error);
5397	return (error);
5398}
5399
5400/*
5401 * Local lock unlock. Unlock all byte ranges that are no longer locked
5402 * by NFSv4. To do this, unlock any subranges of first-->end that
5403 * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5404 * list. This list has all locks for the file held by other
5405 * <clientid, lockowner> tuples. The list is ordered by increasing
5406 * lo_first value, but may have entries that overlap each other, for
5407 * the case of read locks.
5408 */
5409static void
5410nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5411    uint64_t init_end, NFSPROC_T *p)
5412{
5413	struct nfslock *lop;
5414	uint64_t first, end, prevfirst;
5415
5416	first = init_first;
5417	end = init_end;
5418	while (first < init_end) {
5419		/* Loop through all nfs locks, adjusting first and end */
5420		prevfirst = 0;
5421		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5422			KASSERT(prevfirst <= lop->lo_first,
5423			    ("nfsv4 locks out of order"));
5424			KASSERT(lop->lo_first < lop->lo_end,
5425			    ("nfsv4 bogus lock"));
5426			prevfirst = lop->lo_first;
5427			if (first >= lop->lo_first &&
5428			    first < lop->lo_end)
5429				/*
5430				 * Overlaps with initial part, so trim
5431				 * off that initial part by moving first past
5432				 * it.
5433				 */
5434				first = lop->lo_end;
5435			else if (end > lop->lo_first &&
5436			    lop->lo_first > first) {
5437				/*
5438				 * This lock defines the end of the
5439				 * segment to unlock, so set end to the
5440				 * start of it and break out of the loop.
5441				 */
5442				end = lop->lo_first;
5443				break;
5444			}
5445			if (first >= end)
5446				/*
5447				 * There is no segment left to do, so
5448				 * break out of this loop and then exit
5449				 * the outer while() since first will be set
5450				 * to end, which must equal init_end here.
5451				 */
5452				break;
5453		}
5454		if (first < end) {
5455			/* Unlock this segment */
5456			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5457			    NFSLCK_READ, first, end, NULL, p);
5458			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5459			    first, end);
5460		}
5461		/*
5462		 * Now move past this segment and look for any further
5463		 * segment in the range, if there is one.
5464		 */
5465		first = end;
5466		end = init_end;
5467	}
5468}
5469
5470/*
5471 * Do the local lock operation and update the rollback list, as required.
5472 * Perform the rollback and return the error if nfsvno_advlock() fails.
5473 */
5474static int
5475nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5476    uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5477{
5478	struct nfsrollback *rlp;
5479	int error = 0, ltype, oldltype;
5480
5481	if (flags & NFSLCK_WRITE)
5482		ltype = F_WRLCK;
5483	else if (flags & NFSLCK_READ)
5484		ltype = F_RDLCK;
5485	else
5486		ltype = F_UNLCK;
5487	if (oldflags & NFSLCK_WRITE)
5488		oldltype = F_WRLCK;
5489	else if (oldflags & NFSLCK_READ)
5490		oldltype = F_RDLCK;
5491	else
5492		oldltype = F_UNLCK;
5493	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5494		/* nothing to do */
5495		goto out;
5496	error = nfsvno_advlock(vp, ltype, first, end, p);
5497	if (error != 0) {
5498		if (cfp != NULL) {
5499			cfp->cl_clientid.lval[0] = 0;
5500			cfp->cl_clientid.lval[1] = 0;
5501			cfp->cl_first = 0;
5502			cfp->cl_end = NFS64BITSSET;
5503			cfp->cl_flags = NFSLCK_WRITE;
5504			cfp->cl_ownerlen = 5;
5505			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5506		}
5507		nfsrv_locallock_rollback(vp, lfp, p);
5508	} else if (ltype != F_UNLCK) {
5509		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5510		    M_WAITOK);
5511		rlp->rlck_first = first;
5512		rlp->rlck_end = end;
5513		rlp->rlck_type = oldltype;
5514		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5515	}
5516
5517out:
5518	NFSEXITCODE(error);
5519	return (error);
5520}
5521
5522/*
5523 * Roll back local lock changes and free up the rollback list.
5524 */
5525static void
5526nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5527{
5528	struct nfsrollback *rlp, *nrlp;
5529
5530	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5531		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5532		    rlp->rlck_end, p);
5533		free(rlp, M_NFSDROLLBACK);
5534	}
5535	LIST_INIT(&lfp->lf_rollback);
5536}
5537
5538/*
5539 * Update local lock list and delete rollback list (ie now committed to the
5540 * local locks). Most of the work is done by the internal function.
5541 */
5542static void
5543nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5544    uint64_t end)
5545{
5546	struct nfsrollback *rlp, *nrlp;
5547	struct nfslock *new_lop, *other_lop;
5548
5549	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5550	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5551		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5552		    M_WAITOK);
5553	else
5554		other_lop = NULL;
5555	new_lop->lo_flags = flags;
5556	new_lop->lo_first = first;
5557	new_lop->lo_end = end;
5558	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5559	if (new_lop != NULL)
5560		free(new_lop, M_NFSDLOCK);
5561	if (other_lop != NULL)
5562		free(other_lop, M_NFSDLOCK);
5563
5564	/* and get rid of the rollback list */
5565	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5566		free(rlp, M_NFSDROLLBACK);
5567	LIST_INIT(&lfp->lf_rollback);
5568}
5569
5570/*
5571 * Lock the struct nfslockfile for local lock updating.
5572 */
5573static void
5574nfsrv_locklf(struct nfslockfile *lfp)
5575{
5576	int gotlock;
5577
5578	/* lf_usecount ensures *lfp won't be free'd */
5579	lfp->lf_usecount++;
5580	do {
5581		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5582		    NFSSTATEMUTEXPTR, NULL);
5583	} while (gotlock == 0);
5584	lfp->lf_usecount--;
5585}
5586
5587/*
5588 * Unlock the struct nfslockfile after local lock updating.
5589 */
5590static void
5591nfsrv_unlocklf(struct nfslockfile *lfp)
5592{
5593
5594	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5595}
5596
5597/*
5598 * Clear out all state for the NFSv4 server.
5599 * Must be called by a thread that can sleep when no nfsds are running.
5600 */
5601void
5602nfsrv_throwawayallstate(NFSPROC_T *p)
5603{
5604	struct nfsclient *clp, *nclp;
5605	struct nfslockfile *lfp, *nlfp;
5606	int i;
5607
5608	/*
5609	 * For each client, clean out the state and then free the structure.
5610	 */
5611	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
5612		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5613			nfsrv_cleanclient(clp, p);
5614			nfsrv_freedeleglist(&clp->lc_deleg);
5615			nfsrv_freedeleglist(&clp->lc_olddeleg);
5616			free(clp, M_NFSDCLIENT);
5617		}
5618	}
5619
5620	/*
5621	 * Also, free up any remaining lock file structures.
5622	 */
5623	for (i = 0; i < NFSLOCKHASHSIZE; i++) {
5624		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
5625			printf("nfsd unload: fnd a lock file struct\n");
5626			nfsrv_freenfslockfile(lfp);
5627		}
5628	}
5629}
5630
5631/*
5632 * Check the sequence# for the session and slot provided as an argument.
5633 * Also, renew the lease if the session will return NFS_OK.
5634 */
5635int
5636nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
5637    uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
5638    uint32_t *sflagsp, NFSPROC_T *p)
5639{
5640	struct nfsdsession *sep;
5641	struct nfssessionhash *shp;
5642	int error;
5643	SVCXPRT *savxprt;
5644
5645	shp = NFSSESSIONHASH(nd->nd_sessionid);
5646	NFSLOCKSESSION(shp);
5647	sep = nfsrv_findsession(nd->nd_sessionid);
5648	if (sep == NULL) {
5649		NFSUNLOCKSESSION(shp);
5650		return (NFSERR_BADSESSION);
5651	}
5652	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
5653	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
5654	if (error != 0) {
5655		NFSUNLOCKSESSION(shp);
5656		return (error);
5657	}
5658	if (cache_this != 0)
5659		nd->nd_flag |= ND_SAVEREPLY;
5660	/* Renew the lease. */
5661	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
5662	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
5663	nd->nd_flag |= ND_IMPLIEDCLID;
5664
5665	/*
5666	 * If this session handles the backchannel, save the nd_xprt for this
5667	 * RPC, since this is the one being used.
5668	 */
5669	if (sep->sess_cbsess.nfsess_xprt != NULL &&
5670	    (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) {
5671		savxprt = sep->sess_cbsess.nfsess_xprt;
5672		SVC_ACQUIRE(nd->nd_xprt);
5673		nd->nd_xprt->xp_p2 = savxprt->xp_p2;
5674		nd->nd_xprt->xp_idletimeout = 0;	/* Disable timeout. */
5675		sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
5676		SVC_RELEASE(savxprt);
5677	}
5678
5679	*sflagsp = 0;
5680	if (sep->sess_clp->lc_req.nr_client == NULL)
5681		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
5682	NFSUNLOCKSESSION(shp);
5683	if (error == NFSERR_EXPIRED) {
5684		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
5685		error = 0;
5686	} else if (error == NFSERR_ADMINREVOKED) {
5687		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
5688		error = 0;
5689	}
5690	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
5691	return (0);
5692}
5693
5694/*
5695 * Check/set reclaim complete for this session/clientid.
5696 */
5697int
5698nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd)
5699{
5700	struct nfsdsession *sep;
5701	struct nfssessionhash *shp;
5702	int error = 0;
5703
5704	shp = NFSSESSIONHASH(nd->nd_sessionid);
5705	NFSLOCKSTATE();
5706	NFSLOCKSESSION(shp);
5707	sep = nfsrv_findsession(nd->nd_sessionid);
5708	if (sep == NULL) {
5709		NFSUNLOCKSESSION(shp);
5710		NFSUNLOCKSTATE();
5711		return (NFSERR_BADSESSION);
5712	}
5713
5714	/* Check to see if reclaim complete has already happened. */
5715	if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
5716		error = NFSERR_COMPLETEALREADY;
5717	else
5718		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
5719	NFSUNLOCKSESSION(shp);
5720	NFSUNLOCKSTATE();
5721	return (error);
5722}
5723
5724/*
5725 * Cache the reply in a session slot.
5726 */
5727void
5728nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
5729   struct mbuf **m)
5730{
5731	struct nfsdsession *sep;
5732	struct nfssessionhash *shp;
5733
5734	shp = NFSSESSIONHASH(sessionid);
5735	NFSLOCKSESSION(shp);
5736	sep = nfsrv_findsession(sessionid);
5737	if (sep == NULL) {
5738		NFSUNLOCKSESSION(shp);
5739		printf("nfsrv_cache_session: no session\n");
5740		m_freem(*m);
5741		return;
5742	}
5743	nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
5744	NFSUNLOCKSESSION(shp);
5745}
5746
5747/*
5748 * Search for a session that matches the sessionid.
5749 */
5750static struct nfsdsession *
5751nfsrv_findsession(uint8_t *sessionid)
5752{
5753	struct nfsdsession *sep;
5754	struct nfssessionhash *shp;
5755
5756	shp = NFSSESSIONHASH(sessionid);
5757	LIST_FOREACH(sep, &shp->list, sess_hash) {
5758		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
5759			break;
5760	}
5761	return (sep);
5762}
5763
5764/*
5765 * Destroy a session.
5766 */
5767int
5768nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
5769{
5770	int error, samesess;
5771
5772	samesess = 0;
5773	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID)) {
5774		samesess = 1;
5775		if ((nd->nd_flag & ND_LASTOP) == 0)
5776			return (NFSERR_BADSESSION);
5777	}
5778	error = nfsrv_freesession(NULL, sessionid);
5779	if (error == 0 && samesess != 0)
5780		nd->nd_flag &= ~ND_HASSEQUENCE;
5781	return (error);
5782}
5783
5784/*
5785 * Free up a session structure.
5786 */
5787static int
5788nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
5789{
5790	struct nfssessionhash *shp;
5791	int i;
5792
5793	if (sep == NULL) {
5794		shp = NFSSESSIONHASH(sessionid);
5795		NFSLOCKSESSION(shp);
5796		sep = nfsrv_findsession(sessionid);
5797	} else {
5798		shp = NFSSESSIONHASH(sep->sess_sessionid);
5799		NFSLOCKSESSION(shp);
5800	}
5801	if (sep != NULL) {
5802		NFSLOCKSTATE();
5803		sep->sess_refcnt--;
5804		if (sep->sess_refcnt > 0) {
5805			NFSUNLOCKSTATE();
5806			NFSUNLOCKSESSION(shp);
5807			return (0);
5808		}
5809		LIST_REMOVE(sep, sess_hash);
5810		LIST_REMOVE(sep, sess_list);
5811		NFSUNLOCKSTATE();
5812	}
5813	NFSUNLOCKSESSION(shp);
5814	if (sep == NULL)
5815		return (NFSERR_BADSESSION);
5816	for (i = 0; i < NFSV4_SLOTS; i++)
5817		if (sep->sess_slots[i].nfssl_reply != NULL)
5818			m_freem(sep->sess_slots[i].nfssl_reply);
5819	if (sep->sess_cbsess.nfsess_xprt != NULL)
5820		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
5821	free(sep, M_NFSDSESSION);
5822	return (0);
5823}
5824
5825/*
5826 * Free a stateid.
5827 * RFC5661 says that it should fail when there are associated opens, locks
5828 * or delegations. Since stateids represent opens, I don't see how you can
5829 * free an open stateid (it will be free'd when closed), so this function
5830 * only works for lock stateids (freeing the lock_owner) or delegations.
5831 */
5832int
5833nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
5834    NFSPROC_T *p)
5835{
5836	struct nfsclient *clp;
5837	struct nfsstate *stp;
5838	int error;
5839
5840	NFSLOCKSTATE();
5841	/*
5842	 * Look up the stateid
5843	 */
5844	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
5845	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
5846	if (error == 0) {
5847		/* First, check for a delegation. */
5848		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
5849			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
5850			    NFSX_STATEIDOTHER))
5851				break;
5852		}
5853		if (stp != NULL) {
5854			nfsrv_freedeleg(stp);
5855			NFSUNLOCKSTATE();
5856			return (error);
5857		}
5858	}
5859	/* Not a delegation, try for a lock_owner. */
5860	if (error == 0)
5861		error = nfsrv_getstate(clp, stateidp, 0, &stp);
5862	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
5863	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
5864		/* Not a lock_owner stateid. */
5865		error = NFSERR_LOCKSHELD;
5866	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
5867		error = NFSERR_LOCKSHELD;
5868	if (error == 0)
5869		nfsrv_freelockowner(stp, NULL, 0, p);
5870	NFSUNLOCKSTATE();
5871	return (error);
5872}
5873
5874/*
5875 * Generate the xdr for an NFSv4.1 CBSequence Operation.
5876 */
5877static int
5878nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
5879    int dont_replycache, struct nfsdsession **sepp)
5880{
5881	struct nfsdsession *sep;
5882	uint32_t *tl, slotseq = 0;
5883	int maxslot, slotpos;
5884	uint8_t sessionid[NFSX_V4SESSIONID];
5885	int error;
5886
5887	error = nfsv4_getcbsession(clp, sepp);
5888	if (error != 0)
5889		return (error);
5890	sep = *sepp;
5891	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
5892	    &slotseq, sessionid);
5893	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
5894
5895	/* Build the Sequence arguments. */
5896	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
5897	bcopy(sessionid, tl, NFSX_V4SESSIONID);
5898	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
5899	nd->nd_slotseq = tl;
5900	*tl++ = txdr_unsigned(slotseq);
5901	*tl++ = txdr_unsigned(slotpos);
5902	*tl++ = txdr_unsigned(maxslot);
5903	if (dont_replycache == 0)
5904		*tl++ = newnfs_true;
5905	else
5906		*tl++ = newnfs_false;
5907	*tl = 0;			/* No referring call list, for now. */
5908	nd->nd_flag |= ND_HASSEQUENCE;
5909	return (0);
5910}
5911
5912/*
5913 * Get a session for the callback.
5914 */
5915static int
5916nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
5917{
5918	struct nfsdsession *sep;
5919
5920	NFSLOCKSTATE();
5921	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
5922		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
5923			break;
5924	}
5925	if (sep == NULL) {
5926		NFSUNLOCKSTATE();
5927		return (NFSERR_BADSESSION);
5928	}
5929	sep->sess_refcnt++;
5930	*sepp = sep;
5931	NFSUNLOCKSTATE();
5932	return (0);
5933}
5934
5935