nfs_nfsdstate.c revision 358035
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 358035 2020-02-17 19:31:34Z rmacklem $");
30
31#include "opt_inet.h"
32#include "opt_inet6.h"
33#ifndef APPLEKEXT
34#include <fs/nfs/nfsport.h>
35
36struct nfsrv_stablefirst nfsrv_stablefirst;
37int nfsrv_issuedelegs = 0;
38int nfsrv_dolocallocks = 0;
39struct nfsv4lock nfsv4rootfs_lock;
40
41extern int newnfs_numnfsd;
42extern struct nfsstats newnfsstats;
43extern int nfsrv_lease;
44extern struct timeval nfsboottime;
45extern u_int32_t newnfs_true, newnfs_false;
46extern int nfsd_debuglevel;
47NFSV4ROOTLOCKMUTEX;
48NFSSTATESPINLOCK;
49
50SYSCTL_DECL(_vfs_nfsd);
51int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
52TUNABLE_INT("vfs.nfsd.statehashsize", &nfsrv_statehashsize);
53SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
54    &nfsrv_statehashsize, 0,
55    "Size of state hash table set via loader.conf");
56
57int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
58TUNABLE_INT("vfs.nfsd.clienthashsize", &nfsrv_clienthashsize);
59SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
60    &nfsrv_clienthashsize, 0,
61    "Size of client hash table set via loader.conf");
62
63int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
64TUNABLE_INT("vfs.nfsd.fhhashsize", &nfsrv_lockhashsize);
65SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
66    &nfsrv_lockhashsize, 0,
67    "Size of file handle hash table set via loader.conf");
68
69int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
70TUNABLE_INT("vfs.nfsd.sessionhashsize", &nfsrv_sessionhashsize);
71SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
72    &nfsrv_sessionhashsize, 0,
73    "Size of session hash table set via loader.conf");
74
75static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
76TUNABLE_INT("vfs.nfsd.v4statelimit", &nfsrv_v4statelimit);
77SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
78    &nfsrv_v4statelimit, 0,
79    "High water limit for NFSv4 opens+locks+delegations");
80
81static int	nfsrv_writedelegifpos = 0;
82SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
83    &nfsrv_writedelegifpos, 0,
84    "Issue a write delegation for read opens if possible");
85
86static int	nfsrv_allowreadforwriteopen = 1;
87SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
88    &nfsrv_allowreadforwriteopen, 0,
89    "Allow Reads to be done with Write Access StateIDs");
90
91/*
92 * Hash lists for nfs V4.
93 */
94struct nfsclienthashhead	*nfsclienthash;
95struct nfslockhashhead		*nfslockhash;
96struct nfssessionhash		*nfssessionhash;
97#endif	/* !APPLEKEXT */
98
99static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
100static time_t nfsrvboottime;
101static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
102static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
103static int nfsrv_nogsscallback = 0;
104static volatile int nfsrv_writedelegcnt = 0;
105
106/* local functions */
107static void nfsrv_dumpaclient(struct nfsclient *clp,
108    struct nfsd_dumpclients *dumpp);
109static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
110    NFSPROC_T *p);
111static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
112    NFSPROC_T *p);
113static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
114    NFSPROC_T *p);
115static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
116    int cansleep, NFSPROC_T *p);
117static void nfsrv_freenfslock(struct nfslock *lop);
118static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
119static void nfsrv_freedeleg(struct nfsstate *);
120static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
121    u_int32_t flags, struct nfsstate **stpp);
122static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
123    struct nfsstate **stpp);
124static int nfsrv_getlockfh(vnode_t vp, u_short flags,
125    struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
126static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
127    struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
128static void nfsrv_insertlock(struct nfslock *new_lop,
129    struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
130static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
131    struct nfslock **other_lopp, struct nfslockfile *lfp);
132static int nfsrv_getipnumber(u_char *cp);
133static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
134    nfsv4stateid_t *stateidp, int specialid);
135static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
136    u_int32_t flags);
137static int nfsrv_docallback(struct nfsclient *clp, int procnum,
138    nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
139    struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
140static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
141    uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
142static u_int32_t nfsrv_nextclientindex(void);
143static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
144static void nfsrv_markstable(struct nfsclient *clp);
145static void nfsrv_markreclaim(struct nfsclient *clp);
146static int nfsrv_checkstable(struct nfsclient *clp);
147static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
148    vnode *vp, NFSPROC_T *p);
149static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
150    NFSPROC_T *p, vnode_t vp);
151static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
152    struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
153static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
154    struct nfsclient *clp);
155static time_t nfsrv_leaseexpiry(void);
156static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
157static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
158    struct nfsstate *stp, struct nfsrvcache *op);
159static int nfsrv_nootherstate(struct nfsstate *stp);
160static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
161    uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
162static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
163    uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
164static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
165    int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
166    NFSPROC_T *p);
167static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
168    NFSPROC_T *p);
169static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
170    uint64_t first, uint64_t end);
171static void nfsrv_locklf(struct nfslockfile *lfp);
172static void nfsrv_unlocklf(struct nfslockfile *lfp);
173static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
174static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
175static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
176    int dont_replycache, struct nfsdsession **sepp);
177static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
178
179/*
180 * Scan the client list for a match and either return the current one,
181 * create a new entry or return an error.
182 * If returning a non-error, the clp structure must either be linked into
183 * the client list or free'd.
184 */
185APPLESTATIC int
186nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
187    nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
188{
189	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
190	int i, error = 0, ret;
191	struct nfsstate *stp, *tstp;
192#ifdef INET
193	struct sockaddr_in *sin, *rin;
194#endif
195#ifdef INET6
196	struct sockaddr_in6 *sin6, *rin6;
197#endif
198	struct nfsdsession *sep, *nsep;
199	int zapit = 0, gotit, hasstate = 0, igotlock;
200	static u_int64_t confirm_index = 0;
201
202	/*
203	 * Check for state resource limit exceeded.
204	 */
205	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
206		error = NFSERR_RESOURCE;
207		goto out;
208	}
209
210	if (nfsrv_issuedelegs == 0 ||
211	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
212		/*
213		 * Don't do callbacks when delegations are disabled or
214		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
215		 * If establishing a callback connection is attempted
216		 * when a firewall is blocking the callback path, the
217		 * server may wait too long for the connect attempt to
218		 * succeed during the Open. Some clients, such as Linux,
219		 * may timeout and give up on the Open before the server
220		 * replies. Also, since AUTH_GSS callbacks are not
221		 * yet interoperability tested, they might cause the
222		 * server to crap out, if they get past the Init call to
223		 * the client.
224		 */
225		new_clp->lc_program = 0;
226
227	/* Lock out other nfsd threads */
228	NFSLOCKV4ROOTMUTEX();
229	nfsv4_relref(&nfsv4rootfs_lock);
230	do {
231		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
232		    NFSV4ROOTLOCKMUTEXPTR, NULL);
233	} while (!igotlock);
234	NFSUNLOCKV4ROOTMUTEX();
235
236	/*
237	 * Search for a match in the client list.
238	 */
239	gotit = i = 0;
240	while (i < nfsrv_clienthashsize && !gotit) {
241	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
242		if (new_clp->lc_idlen == clp->lc_idlen &&
243		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
244			gotit = 1;
245			break;
246		}
247	    }
248	    if (gotit == 0)
249		i++;
250	}
251	if (!gotit ||
252	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
253		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
254			/*
255			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
256			 * client is trying to update a confirmed clientid.
257			 */
258			NFSLOCKV4ROOTMUTEX();
259			nfsv4_unlock(&nfsv4rootfs_lock, 1);
260			NFSUNLOCKV4ROOTMUTEX();
261			confirmp->lval[1] = 0;
262			error = NFSERR_NOENT;
263			goto out;
264		}
265		/*
266		 * Get rid of the old one.
267		 */
268		if (i != nfsrv_clienthashsize) {
269			LIST_REMOVE(clp, lc_hash);
270			nfsrv_cleanclient(clp, p);
271			nfsrv_freedeleglist(&clp->lc_deleg);
272			nfsrv_freedeleglist(&clp->lc_olddeleg);
273			zapit = 1;
274		}
275		/*
276		 * Add it after assigning a client id to it.
277		 */
278		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
279		if ((nd->nd_flag & ND_NFSV41) != 0)
280			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
281			    ++confirm_index;
282		else
283			confirmp->qval = new_clp->lc_confirm.qval =
284			    ++confirm_index;
285		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
286		    (u_int32_t)nfsrvboottime;
287		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
288		    nfsrv_nextclientindex();
289		new_clp->lc_stateindex = 0;
290		new_clp->lc_statemaxindex = 0;
291		new_clp->lc_cbref = 0;
292		new_clp->lc_expiry = nfsrv_leaseexpiry();
293		LIST_INIT(&new_clp->lc_open);
294		LIST_INIT(&new_clp->lc_deleg);
295		LIST_INIT(&new_clp->lc_olddeleg);
296		LIST_INIT(&new_clp->lc_session);
297		for (i = 0; i < nfsrv_statehashsize; i++)
298			LIST_INIT(&new_clp->lc_stateid[i]);
299		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
300		    lc_hash);
301		newnfsstats.srvclients++;
302		nfsrv_openpluslock++;
303		nfsrv_clients++;
304		NFSLOCKV4ROOTMUTEX();
305		nfsv4_unlock(&nfsv4rootfs_lock, 1);
306		NFSUNLOCKV4ROOTMUTEX();
307		if (zapit)
308			nfsrv_zapclient(clp, p);
309		*new_clpp = NULL;
310		goto out;
311	}
312
313	/*
314	 * Now, handle the cases where the id is already issued.
315	 */
316	if (nfsrv_notsamecredname(nd, clp)) {
317	    /*
318	     * Check to see if there is expired state that should go away.
319	     */
320	    if (clp->lc_expiry < NFSD_MONOSEC &&
321	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
322		nfsrv_cleanclient(clp, p);
323		nfsrv_freedeleglist(&clp->lc_deleg);
324	    }
325
326	    /*
327	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
328	     * RFC3530 Sec. 8.1.2 last para.
329	     */
330	    if (!LIST_EMPTY(&clp->lc_deleg)) {
331		hasstate = 1;
332	    } else if (LIST_EMPTY(&clp->lc_open)) {
333		hasstate = 0;
334	    } else {
335		hasstate = 0;
336		/* Look for an Open on the OpenOwner */
337		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
338		    if (!LIST_EMPTY(&stp->ls_open)) {
339			hasstate = 1;
340			break;
341		    }
342		}
343	    }
344	    if (hasstate) {
345		/*
346		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
347		 * filling out the correct ipaddr and portnum.
348		 */
349		switch (clp->lc_req.nr_nam->sa_family) {
350#ifdef INET
351		case AF_INET:
352			sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam;
353			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
354			sin->sin_addr.s_addr = rin->sin_addr.s_addr;
355			sin->sin_port = rin->sin_port;
356			break;
357#endif
358#ifdef INET6
359		case AF_INET6:
360			sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam;
361			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
362			sin6->sin6_addr = rin6->sin6_addr;
363			sin6->sin6_port = rin6->sin6_port;
364			break;
365#endif
366		}
367		NFSLOCKV4ROOTMUTEX();
368		nfsv4_unlock(&nfsv4rootfs_lock, 1);
369		NFSUNLOCKV4ROOTMUTEX();
370		error = NFSERR_CLIDINUSE;
371		goto out;
372	    }
373	}
374
375	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
376		/*
377		 * If the verifier has changed, the client has rebooted
378		 * and a new client id is issued. The old state info
379		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
380		 */
381		LIST_REMOVE(clp, lc_hash);
382
383		/* Get rid of all sessions on this clientid. */
384		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) {
385			ret = nfsrv_freesession(sep, NULL);
386			if (ret != 0)
387				printf("nfsrv_setclient: verifier changed free"
388				    " session failed=%d\n", ret);
389		}
390
391		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
392		if ((nd->nd_flag & ND_NFSV41) != 0)
393			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
394			    ++confirm_index;
395		else
396			confirmp->qval = new_clp->lc_confirm.qval =
397			    ++confirm_index;
398		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
399		    nfsrvboottime;
400		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
401		    nfsrv_nextclientindex();
402		new_clp->lc_stateindex = 0;
403		new_clp->lc_statemaxindex = 0;
404		new_clp->lc_cbref = 0;
405		new_clp->lc_expiry = nfsrv_leaseexpiry();
406
407		/*
408		 * Save the state until confirmed.
409		 */
410		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
411		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
412			tstp->ls_clp = new_clp;
413		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
414		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
415			tstp->ls_clp = new_clp;
416		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
417		    ls_list);
418		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
419			tstp->ls_clp = new_clp;
420		for (i = 0; i < nfsrv_statehashsize; i++) {
421			LIST_NEWHEAD(&new_clp->lc_stateid[i],
422			    &clp->lc_stateid[i], ls_hash);
423			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
424				tstp->ls_clp = new_clp;
425		}
426		LIST_INIT(&new_clp->lc_session);
427		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
428		    lc_hash);
429		newnfsstats.srvclients++;
430		nfsrv_openpluslock++;
431		nfsrv_clients++;
432		NFSLOCKV4ROOTMUTEX();
433		nfsv4_unlock(&nfsv4rootfs_lock, 1);
434		NFSUNLOCKV4ROOTMUTEX();
435
436		/*
437		 * Must wait until any outstanding callback on the old clp
438		 * completes.
439		 */
440		NFSLOCKSTATE();
441		while (clp->lc_cbref) {
442			clp->lc_flags |= LCL_WAKEUPWANTED;
443			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
444			    "nfsd clp", 10 * hz);
445		}
446		NFSUNLOCKSTATE();
447		nfsrv_zapclient(clp, p);
448		*new_clpp = NULL;
449		goto out;
450	}
451
452	/* For NFSv4.1, mark that we found a confirmed clientid. */
453	if ((nd->nd_flag & ND_NFSV41) != 0) {
454		clientidp->lval[0] = clp->lc_clientid.lval[0];
455		clientidp->lval[1] = clp->lc_clientid.lval[1];
456		confirmp->lval[0] = 0;	/* Ignored by client */
457		confirmp->lval[1] = 1;
458	} else {
459		/*
460		 * id and verifier match, so update the net address info
461		 * and get rid of any existing callback authentication
462		 * handle, so a new one will be acquired.
463		 */
464		LIST_REMOVE(clp, lc_hash);
465		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
466		new_clp->lc_expiry = nfsrv_leaseexpiry();
467		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
468		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
469		    clp->lc_clientid.lval[0];
470		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
471		    clp->lc_clientid.lval[1];
472		new_clp->lc_delegtime = clp->lc_delegtime;
473		new_clp->lc_stateindex = clp->lc_stateindex;
474		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
475		new_clp->lc_cbref = 0;
476		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
477		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
478			tstp->ls_clp = new_clp;
479		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
480		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
481			tstp->ls_clp = new_clp;
482		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
483		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
484			tstp->ls_clp = new_clp;
485		for (i = 0; i < nfsrv_statehashsize; i++) {
486			LIST_NEWHEAD(&new_clp->lc_stateid[i],
487			    &clp->lc_stateid[i], ls_hash);
488			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
489				tstp->ls_clp = new_clp;
490		}
491		LIST_INIT(&new_clp->lc_session);
492		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
493		    lc_hash);
494		newnfsstats.srvclients++;
495		nfsrv_openpluslock++;
496		nfsrv_clients++;
497	}
498	NFSLOCKV4ROOTMUTEX();
499	nfsv4_unlock(&nfsv4rootfs_lock, 1);
500	NFSUNLOCKV4ROOTMUTEX();
501
502	if ((nd->nd_flag & ND_NFSV41) == 0) {
503		/*
504		 * Must wait until any outstanding callback on the old clp
505		 * completes.
506		 */
507		NFSLOCKSTATE();
508		while (clp->lc_cbref) {
509			clp->lc_flags |= LCL_WAKEUPWANTED;
510			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
511			    "nfsdclp", 10 * hz);
512		}
513		NFSUNLOCKSTATE();
514		nfsrv_zapclient(clp, p);
515		*new_clpp = NULL;
516	}
517
518out:
519	NFSEXITCODE2(error, nd);
520	return (error);
521}
522
523/*
524 * Check to see if the client id exists and optionally confirm it.
525 */
526APPLESTATIC int
527nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
528    struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
529    struct nfsrv_descript *nd, NFSPROC_T *p)
530{
531	struct nfsclient *clp;
532	struct nfsstate *stp;
533	int i;
534	struct nfsclienthashhead *hp;
535	int error = 0, igotlock, doneok;
536	struct nfssessionhash *shp;
537	struct nfsdsession *sep;
538	uint64_t sessid[2];
539	static uint64_t next_sess = 0;
540
541	if (clpp)
542		*clpp = NULL;
543	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
544	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
545		error = NFSERR_STALECLIENTID;
546		goto out;
547	}
548
549	/*
550	 * If called with opflags == CLOPS_RENEW, the State Lock is
551	 * already held. Otherwise, we need to get either that or,
552	 * for the case of Confirm, lock out the nfsd threads.
553	 */
554	if (opflags & CLOPS_CONFIRM) {
555		NFSLOCKV4ROOTMUTEX();
556		nfsv4_relref(&nfsv4rootfs_lock);
557		do {
558			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
559			    NFSV4ROOTLOCKMUTEXPTR, NULL);
560		} while (!igotlock);
561		/*
562		 * Create a new sessionid here, since we need to do it where
563		 * there is a mutex held to serialize update of next_sess.
564		 */
565		if ((nd->nd_flag & ND_NFSV41) != 0) {
566			sessid[0] = ++next_sess;
567			sessid[1] = clientid.qval;
568		}
569		NFSUNLOCKV4ROOTMUTEX();
570	} else if (opflags != CLOPS_RENEW) {
571		NFSLOCKSTATE();
572	}
573
574	/* For NFSv4.1, the clp is acquired from the associated session. */
575	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
576	    opflags == CLOPS_RENEW) {
577		clp = NULL;
578		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
579			shp = NFSSESSIONHASH(nd->nd_sessionid);
580			NFSLOCKSESSION(shp);
581			sep = nfsrv_findsession(nd->nd_sessionid);
582			if (sep != NULL)
583				clp = sep->sess_clp;
584			NFSUNLOCKSESSION(shp);
585		}
586	} else {
587		hp = NFSCLIENTHASH(clientid);
588		LIST_FOREACH(clp, hp, lc_hash) {
589			if (clp->lc_clientid.lval[1] == clientid.lval[1])
590				break;
591		}
592	}
593	if (clp == NULL) {
594		if (opflags & CLOPS_CONFIRM)
595			error = NFSERR_STALECLIENTID;
596		else
597			error = NFSERR_EXPIRED;
598	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
599		/*
600		 * If marked admin revoked, just return the error.
601		 */
602		error = NFSERR_ADMINREVOKED;
603	}
604	if (error) {
605		if (opflags & CLOPS_CONFIRM) {
606			NFSLOCKV4ROOTMUTEX();
607			nfsv4_unlock(&nfsv4rootfs_lock, 1);
608			NFSUNLOCKV4ROOTMUTEX();
609		} else if (opflags != CLOPS_RENEW) {
610			NFSUNLOCKSTATE();
611		}
612		goto out;
613	}
614
615	/*
616	 * Perform any operations specified by the opflags.
617	 */
618	if (opflags & CLOPS_CONFIRM) {
619		if (((nd->nd_flag & ND_NFSV41) != 0 &&
620		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
621		    ((nd->nd_flag & ND_NFSV41) == 0 &&
622		     clp->lc_confirm.qval != confirm.qval))
623			error = NFSERR_STALECLIENTID;
624		else if (nfsrv_notsamecredname(nd, clp))
625			error = NFSERR_CLIDINUSE;
626
627		if (!error) {
628		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
629			LCL_NEEDSCONFIRM) {
630			/*
631			 * Hang onto the delegations (as old delegations)
632			 * for an Open with CLAIM_DELEGATE_PREV unless in
633			 * grace, but get rid of the rest of the state.
634			 */
635			nfsrv_cleanclient(clp, p);
636			nfsrv_freedeleglist(&clp->lc_olddeleg);
637			if (nfsrv_checkgrace(nd, clp, 0)) {
638			    /* In grace, so just delete delegations */
639			    nfsrv_freedeleglist(&clp->lc_deleg);
640			} else {
641			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
642				stp->ls_flags |= NFSLCK_OLDDELEG;
643			    clp->lc_delegtime = NFSD_MONOSEC +
644				nfsrv_lease + NFSRV_LEASEDELTA;
645			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
646				ls_list);
647			}
648			if ((nd->nd_flag & ND_NFSV41) != 0)
649			    clp->lc_program = cbprogram;
650		    }
651		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
652		    if (clp->lc_program)
653			clp->lc_flags |= LCL_NEEDSCBNULL;
654		    /* For NFSv4.1, link the session onto the client. */
655		    if (nsep != NULL) {
656			/* Hold a reference on the xprt for a backchannel. */
657			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
658			    != 0) {
659			    if (clp->lc_req.nr_client == NULL)
660				clp->lc_req.nr_client = (struct __rpc_client *)
661				    clnt_bck_create(nd->nd_xprt->xp_socket,
662				    cbprogram, NFSV4_CBVERS);
663			    if (clp->lc_req.nr_client != NULL) {
664				SVC_ACQUIRE(nd->nd_xprt);
665				nd->nd_xprt->xp_p2 =
666				    clp->lc_req.nr_client->cl_private;
667				/* Disable idle timeout. */
668				nd->nd_xprt->xp_idletimeout = 0;
669				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
670			    } else
671				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
672			}
673			NFSBCOPY(sessid, nsep->sess_sessionid,
674			    NFSX_V4SESSIONID);
675			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
676			    NFSX_V4SESSIONID);
677			shp = NFSSESSIONHASH(nsep->sess_sessionid);
678			NFSLOCKSTATE();
679			NFSLOCKSESSION(shp);
680			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
681			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
682			nsep->sess_clp = clp;
683			NFSUNLOCKSESSION(shp);
684			NFSUNLOCKSTATE();
685		    }
686		}
687	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
688		error = NFSERR_EXPIRED;
689	}
690
691	/*
692	 * If called by the Renew Op, we must check the principal.
693	 */
694	if (!error && (opflags & CLOPS_RENEWOP)) {
695	    if (nfsrv_notsamecredname(nd, clp)) {
696		doneok = 0;
697		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
698		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
699			if ((stp->ls_flags & NFSLCK_OPEN) &&
700			    stp->ls_uid == nd->nd_cred->cr_uid) {
701				doneok = 1;
702				break;
703			}
704		    }
705		}
706		if (!doneok)
707			error = NFSERR_ACCES;
708	    }
709	    if (!error && (clp->lc_flags & LCL_CBDOWN))
710		error = NFSERR_CBPATHDOWN;
711	}
712	if ((!error || error == NFSERR_CBPATHDOWN) &&
713	     (opflags & CLOPS_RENEW)) {
714		clp->lc_expiry = nfsrv_leaseexpiry();
715	}
716	if (opflags & CLOPS_CONFIRM) {
717		NFSLOCKV4ROOTMUTEX();
718		nfsv4_unlock(&nfsv4rootfs_lock, 1);
719		NFSUNLOCKV4ROOTMUTEX();
720	} else if (opflags != CLOPS_RENEW) {
721		NFSUNLOCKSTATE();
722	}
723	if (clpp)
724		*clpp = clp;
725
726out:
727	NFSEXITCODE2(error, nd);
728	return (error);
729}
730
731/*
732 * Perform the NFSv4.1 destroy clientid.
733 */
734int
735nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
736{
737	struct nfsclient *clp;
738	struct nfsclienthashhead *hp;
739	int error = 0, i, igotlock;
740
741	if (nfsrvboottime != clientid.lval[0]) {
742		error = NFSERR_STALECLIENTID;
743		goto out;
744	}
745
746	/* Lock out other nfsd threads */
747	NFSLOCKV4ROOTMUTEX();
748	nfsv4_relref(&nfsv4rootfs_lock);
749	do {
750		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
751		    NFSV4ROOTLOCKMUTEXPTR, NULL);
752	} while (igotlock == 0);
753	NFSUNLOCKV4ROOTMUTEX();
754
755	hp = NFSCLIENTHASH(clientid);
756	LIST_FOREACH(clp, hp, lc_hash) {
757		if (clp->lc_clientid.lval[1] == clientid.lval[1])
758			break;
759	}
760	if (clp == NULL) {
761		NFSLOCKV4ROOTMUTEX();
762		nfsv4_unlock(&nfsv4rootfs_lock, 1);
763		NFSUNLOCKV4ROOTMUTEX();
764		/* Just return ok, since it is gone. */
765		goto out;
766	}
767
768	/* Scan for state on the clientid. */
769	for (i = 0; i < nfsrv_statehashsize; i++)
770		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
771			NFSLOCKV4ROOTMUTEX();
772			nfsv4_unlock(&nfsv4rootfs_lock, 1);
773			NFSUNLOCKV4ROOTMUTEX();
774			error = NFSERR_CLIENTIDBUSY;
775			goto out;
776		}
777	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
778		NFSLOCKV4ROOTMUTEX();
779		nfsv4_unlock(&nfsv4rootfs_lock, 1);
780		NFSUNLOCKV4ROOTMUTEX();
781		error = NFSERR_CLIENTIDBUSY;
782		goto out;
783	}
784
785	/* Destroy the clientid and return ok. */
786	nfsrv_cleanclient(clp, p);
787	nfsrv_freedeleglist(&clp->lc_deleg);
788	nfsrv_freedeleglist(&clp->lc_olddeleg);
789	LIST_REMOVE(clp, lc_hash);
790	NFSLOCKV4ROOTMUTEX();
791	nfsv4_unlock(&nfsv4rootfs_lock, 1);
792	NFSUNLOCKV4ROOTMUTEX();
793	nfsrv_zapclient(clp, p);
794out:
795	NFSEXITCODE2(error, nd);
796	return (error);
797}
798
799/*
800 * Called from the new nfssvc syscall to admin revoke a clientid.
801 * Returns 0 for success, error otherwise.
802 */
803APPLESTATIC int
804nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
805{
806	struct nfsclient *clp = NULL;
807	int i, error = 0;
808	int gotit, igotlock;
809
810	/*
811	 * First, lock out the nfsd so that state won't change while the
812	 * revocation record is being written to the stable storage restart
813	 * file.
814	 */
815	NFSLOCKV4ROOTMUTEX();
816	do {
817		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
818		    NFSV4ROOTLOCKMUTEXPTR, NULL);
819	} while (!igotlock);
820	NFSUNLOCKV4ROOTMUTEX();
821
822	/*
823	 * Search for a match in the client list.
824	 */
825	gotit = i = 0;
826	while (i < nfsrv_clienthashsize && !gotit) {
827	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
828		if (revokep->nclid_idlen == clp->lc_idlen &&
829		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
830			gotit = 1;
831			break;
832		}
833	    }
834	    i++;
835	}
836	if (!gotit) {
837		NFSLOCKV4ROOTMUTEX();
838		nfsv4_unlock(&nfsv4rootfs_lock, 0);
839		NFSUNLOCKV4ROOTMUTEX();
840		error = EPERM;
841		goto out;
842	}
843
844	/*
845	 * Now, write out the revocation record
846	 */
847	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
848	nfsrv_backupstable();
849
850	/*
851	 * and clear out the state, marking the clientid revoked.
852	 */
853	clp->lc_flags &= ~LCL_CALLBACKSON;
854	clp->lc_flags |= LCL_ADMINREVOKED;
855	nfsrv_cleanclient(clp, p);
856	nfsrv_freedeleglist(&clp->lc_deleg);
857	nfsrv_freedeleglist(&clp->lc_olddeleg);
858	NFSLOCKV4ROOTMUTEX();
859	nfsv4_unlock(&nfsv4rootfs_lock, 0);
860	NFSUNLOCKV4ROOTMUTEX();
861
862out:
863	NFSEXITCODE(error);
864	return (error);
865}
866
867/*
868 * Dump out stats for all clients. Called from nfssvc(2), that is used
869 * newnfsstats.
870 */
871APPLESTATIC void
872nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
873{
874	struct nfsclient *clp;
875	int i = 0, cnt = 0;
876
877	/*
878	 * First, get a reference on the nfsv4rootfs_lock so that an
879	 * exclusive lock cannot be acquired while dumping the clients.
880	 */
881	NFSLOCKV4ROOTMUTEX();
882	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
883	NFSUNLOCKV4ROOTMUTEX();
884	NFSLOCKSTATE();
885	/*
886	 * Rattle through the client lists until done.
887	 */
888	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
889	    clp = LIST_FIRST(&nfsclienthash[i]);
890	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
891		nfsrv_dumpaclient(clp, &dumpp[cnt]);
892		cnt++;
893		clp = LIST_NEXT(clp, lc_hash);
894	    }
895	    i++;
896	}
897	if (cnt < maxcnt)
898	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
899	NFSUNLOCKSTATE();
900	NFSLOCKV4ROOTMUTEX();
901	nfsv4_relref(&nfsv4rootfs_lock);
902	NFSUNLOCKV4ROOTMUTEX();
903}
904
905/*
906 * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
907 */
908static void
909nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
910{
911	struct nfsstate *stp, *openstp, *lckownstp;
912	struct nfslock *lop;
913	sa_family_t af;
914#ifdef INET
915	struct sockaddr_in *rin;
916#endif
917#ifdef INET6
918	struct sockaddr_in6 *rin6;
919#endif
920
921	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
922	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
923	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
924	dumpp->ndcl_flags = clp->lc_flags;
925	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
926	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
927	af = clp->lc_req.nr_nam->sa_family;
928	dumpp->ndcl_addrfam = af;
929	switch (af) {
930#ifdef INET
931	case AF_INET:
932		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
933		dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr;
934		break;
935#endif
936#ifdef INET6
937	case AF_INET6:
938		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
939		dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr;
940		break;
941#endif
942	}
943
944	/*
945	 * Now, scan the state lists and total up the opens and locks.
946	 */
947	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
948	    dumpp->ndcl_nopenowners++;
949	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
950		dumpp->ndcl_nopens++;
951		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
952		    dumpp->ndcl_nlockowners++;
953		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
954			dumpp->ndcl_nlocks++;
955		    }
956		}
957	    }
958	}
959
960	/*
961	 * and the delegation lists.
962	 */
963	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
964	    dumpp->ndcl_ndelegs++;
965	}
966	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
967	    dumpp->ndcl_nolddelegs++;
968	}
969}
970
971/*
972 * Dump out lock stats for a file.
973 */
974APPLESTATIC void
975nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
976    NFSPROC_T *p)
977{
978	struct nfsstate *stp;
979	struct nfslock *lop;
980	int cnt = 0;
981	struct nfslockfile *lfp;
982	sa_family_t af;
983#ifdef INET
984	struct sockaddr_in *rin;
985#endif
986#ifdef INET6
987	struct sockaddr_in6 *rin6;
988#endif
989	int ret;
990	fhandle_t nfh;
991
992	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
993	/*
994	 * First, get a reference on the nfsv4rootfs_lock so that an
995	 * exclusive lock on it cannot be acquired while dumping the locks.
996	 */
997	NFSLOCKV4ROOTMUTEX();
998	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
999	NFSUNLOCKV4ROOTMUTEX();
1000	NFSLOCKSTATE();
1001	if (!ret)
1002		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1003	if (ret) {
1004		ldumpp[0].ndlck_clid.nclid_idlen = 0;
1005		NFSUNLOCKSTATE();
1006		NFSLOCKV4ROOTMUTEX();
1007		nfsv4_relref(&nfsv4rootfs_lock);
1008		NFSUNLOCKV4ROOTMUTEX();
1009		return;
1010	}
1011
1012	/*
1013	 * For each open share on file, dump it out.
1014	 */
1015	stp = LIST_FIRST(&lfp->lf_open);
1016	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1017		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1018		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1019		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1020		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1021		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1022		ldumpp[cnt].ndlck_owner.nclid_idlen =
1023		    stp->ls_openowner->ls_ownerlen;
1024		NFSBCOPY(stp->ls_openowner->ls_owner,
1025		    ldumpp[cnt].ndlck_owner.nclid_id,
1026		    stp->ls_openowner->ls_ownerlen);
1027		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1028		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1029		    stp->ls_clp->lc_idlen);
1030		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1031		ldumpp[cnt].ndlck_addrfam = af;
1032		switch (af) {
1033#ifdef INET
1034		case AF_INET:
1035			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1036			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1037			break;
1038#endif
1039#ifdef INET6
1040		case AF_INET6:
1041			rin6 = (struct sockaddr_in6 *)
1042			    stp->ls_clp->lc_req.nr_nam;
1043			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1044			break;
1045#endif
1046		}
1047		stp = LIST_NEXT(stp, ls_file);
1048		cnt++;
1049	}
1050
1051	/*
1052	 * and all locks.
1053	 */
1054	lop = LIST_FIRST(&lfp->lf_lock);
1055	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1056		stp = lop->lo_stp;
1057		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1058		ldumpp[cnt].ndlck_first = lop->lo_first;
1059		ldumpp[cnt].ndlck_end = lop->lo_end;
1060		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1061		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1062		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1063		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1064		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1065		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1066		    stp->ls_ownerlen);
1067		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1068		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1069		    stp->ls_clp->lc_idlen);
1070		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1071		ldumpp[cnt].ndlck_addrfam = af;
1072		switch (af) {
1073#ifdef INET
1074		case AF_INET:
1075			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1076			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1077			break;
1078#endif
1079#ifdef INET6
1080		case AF_INET6:
1081			rin6 = (struct sockaddr_in6 *)
1082			    stp->ls_clp->lc_req.nr_nam;
1083			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1084			break;
1085#endif
1086		}
1087		lop = LIST_NEXT(lop, lo_lckfile);
1088		cnt++;
1089	}
1090
1091	/*
1092	 * and the delegations.
1093	 */
1094	stp = LIST_FIRST(&lfp->lf_deleg);
1095	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1096		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1097		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1098		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1099		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1100		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1101		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1102		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1103		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1104		    stp->ls_clp->lc_idlen);
1105		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1106		ldumpp[cnt].ndlck_addrfam = af;
1107		switch (af) {
1108#ifdef INET
1109		case AF_INET:
1110			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1111			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1112			break;
1113#endif
1114#ifdef INET6
1115		case AF_INET6:
1116			rin6 = (struct sockaddr_in6 *)
1117			    stp->ls_clp->lc_req.nr_nam;
1118			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1119			break;
1120#endif
1121		}
1122		stp = LIST_NEXT(stp, ls_file);
1123		cnt++;
1124	}
1125
1126	/*
1127	 * If list isn't full, mark end of list by setting the client name
1128	 * to zero length.
1129	 */
1130	if (cnt < maxcnt)
1131		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1132	NFSUNLOCKSTATE();
1133	NFSLOCKV4ROOTMUTEX();
1134	nfsv4_relref(&nfsv4rootfs_lock);
1135	NFSUNLOCKV4ROOTMUTEX();
1136}
1137
1138/*
1139 * Server timer routine. It can scan any linked list, so long
1140 * as it holds the spin/mutex lock and there is no exclusive lock on
1141 * nfsv4rootfs_lock.
1142 * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1143 *  to do this from a callout, since the spin locks work. For
1144 *  Darwin, I'm not sure what will work correctly yet.)
1145 * Should be called once per second.
1146 */
1147APPLESTATIC void
1148nfsrv_servertimer(void)
1149{
1150	struct nfsclient *clp, *nclp;
1151	struct nfsstate *stp, *nstp;
1152	int got_ref, i;
1153
1154	/*
1155	 * Make sure nfsboottime is set. This is used by V3 as well
1156	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1157	 * only used by the V4 server for leases.
1158	 */
1159	if (nfsboottime.tv_sec == 0)
1160		NFSSETBOOTTIME(nfsboottime);
1161
1162	/*
1163	 * If server hasn't started yet, just return.
1164	 */
1165	NFSLOCKSTATE();
1166	if (nfsrv_stablefirst.nsf_eograce == 0) {
1167		NFSUNLOCKSTATE();
1168		return;
1169	}
1170	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1171		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1172		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1173			nfsrv_stablefirst.nsf_flags |=
1174			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1175		NFSUNLOCKSTATE();
1176		return;
1177	}
1178
1179	/*
1180	 * Try and get a reference count on the nfsv4rootfs_lock so that
1181	 * no nfsd thread can acquire an exclusive lock on it before this
1182	 * call is done. If it is already exclusively locked, just return.
1183	 */
1184	NFSLOCKV4ROOTMUTEX();
1185	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1186	NFSUNLOCKV4ROOTMUTEX();
1187	if (got_ref == 0) {
1188		NFSUNLOCKSTATE();
1189		return;
1190	}
1191
1192	/*
1193	 * For each client...
1194	 */
1195	for (i = 0; i < nfsrv_clienthashsize; i++) {
1196	    clp = LIST_FIRST(&nfsclienthash[i]);
1197	    while (clp != LIST_END(&nfsclienthash[i])) {
1198		nclp = LIST_NEXT(clp, lc_hash);
1199		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1200		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1201			 && ((LIST_EMPTY(&clp->lc_deleg)
1202			      && LIST_EMPTY(&clp->lc_open)) ||
1203			     nfsrv_clients > nfsrv_clienthighwater)) ||
1204			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1205			(clp->lc_expiry < NFSD_MONOSEC &&
1206			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1207			/*
1208			 * Lease has expired several nfsrv_lease times ago:
1209			 * PLUS
1210			 *    - no state is associated with it
1211			 *    OR
1212			 *    - above high water mark for number of clients
1213			 *      (nfsrv_clienthighwater should be large enough
1214			 *       that this only occurs when clients fail to
1215			 *       use the same nfs_client_id4.id. Maybe somewhat
1216			 *       higher that the maximum number of clients that
1217			 *       will mount this server?)
1218			 * OR
1219			 * Lease has expired a very long time ago
1220			 * OR
1221			 * Lease has expired PLUS the number of opens + locks
1222			 * has exceeded 90% of capacity
1223			 *
1224			 * --> Mark for expiry. The actual expiry will be done
1225			 *     by an nfsd sometime soon.
1226			 */
1227			clp->lc_flags |= LCL_EXPIREIT;
1228			nfsrv_stablefirst.nsf_flags |=
1229			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1230		    } else {
1231			/*
1232			 * If there are no opens, increment no open tick cnt
1233			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1234			 * otherwise, if there is an open, reset no open time
1235			 * Hopefully, this will avoid excessive re-creation
1236			 * of open owners and subsequent open confirms.
1237			 */
1238			stp = LIST_FIRST(&clp->lc_open);
1239			while (stp != LIST_END(&clp->lc_open)) {
1240				nstp = LIST_NEXT(stp, ls_list);
1241				if (LIST_EMPTY(&stp->ls_open)) {
1242					stp->ls_noopens++;
1243					if (stp->ls_noopens > NFSNOOPEN ||
1244					    (nfsrv_openpluslock * 2) >
1245					    nfsrv_v4statelimit)
1246						nfsrv_stablefirst.nsf_flags |=
1247							NFSNSF_NOOPENS;
1248				} else {
1249					stp->ls_noopens = 0;
1250				}
1251				stp = nstp;
1252			}
1253		    }
1254		}
1255		clp = nclp;
1256	    }
1257	}
1258	NFSUNLOCKSTATE();
1259	NFSLOCKV4ROOTMUTEX();
1260	nfsv4_relref(&nfsv4rootfs_lock);
1261	NFSUNLOCKV4ROOTMUTEX();
1262}
1263
1264/*
1265 * The following set of functions free up the various data structures.
1266 */
1267/*
1268 * Clear out all open/lock state related to this nfsclient.
1269 * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1270 * there are no other active nfsd threads.
1271 */
1272APPLESTATIC void
1273nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1274{
1275	struct nfsstate *stp, *nstp;
1276	struct nfsdsession *sep, *nsep;
1277
1278	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1279		nfsrv_freeopenowner(stp, 1, p);
1280	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1281		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1282			(void)nfsrv_freesession(sep, NULL);
1283}
1284
1285/*
1286 * Free a client that has been cleaned. It should also already have been
1287 * removed from the lists.
1288 * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1289 *  softclock interrupts are enabled.)
1290 */
1291APPLESTATIC void
1292nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1293{
1294
1295#ifdef notyet
1296	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1297	     (LCL_GSS | LCL_CALLBACKSON) &&
1298	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1299	    clp->lc_handlelen > 0) {
1300		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1301		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1302		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1303			NULL, 0, NULL, NULL, NULL, p);
1304	}
1305#endif
1306	newnfs_disconnect(&clp->lc_req);
1307	NFSSOCKADDRFREE(clp->lc_req.nr_nam);
1308	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1309	free(clp->lc_stateid, M_NFSDCLIENT);
1310	free(clp, M_NFSDCLIENT);
1311	NFSLOCKSTATE();
1312	newnfsstats.srvclients--;
1313	nfsrv_openpluslock--;
1314	nfsrv_clients--;
1315	NFSUNLOCKSTATE();
1316}
1317
1318/*
1319 * Free a list of delegation state structures.
1320 * (This function will also free all nfslockfile structures that no
1321 *  longer have associated state.)
1322 */
1323APPLESTATIC void
1324nfsrv_freedeleglist(struct nfsstatehead *sthp)
1325{
1326	struct nfsstate *stp, *nstp;
1327
1328	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1329		nfsrv_freedeleg(stp);
1330	}
1331	LIST_INIT(sthp);
1332}
1333
1334/*
1335 * Free up a delegation.
1336 */
1337static void
1338nfsrv_freedeleg(struct nfsstate *stp)
1339{
1340	struct nfslockfile *lfp;
1341
1342	LIST_REMOVE(stp, ls_hash);
1343	LIST_REMOVE(stp, ls_list);
1344	LIST_REMOVE(stp, ls_file);
1345	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1346		nfsrv_writedelegcnt--;
1347	lfp = stp->ls_lfp;
1348	if (LIST_EMPTY(&lfp->lf_open) &&
1349	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1350	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1351	    lfp->lf_usecount == 0 &&
1352	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1353		nfsrv_freenfslockfile(lfp);
1354	FREE((caddr_t)stp, M_NFSDSTATE);
1355	newnfsstats.srvdelegates--;
1356	nfsrv_openpluslock--;
1357	nfsrv_delegatecnt--;
1358}
1359
1360/*
1361 * This function frees an open owner and all associated opens.
1362 */
1363static void
1364nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1365{
1366	struct nfsstate *nstp, *tstp;
1367
1368	LIST_REMOVE(stp, ls_list);
1369	/*
1370	 * Now, free all associated opens.
1371	 */
1372	nstp = LIST_FIRST(&stp->ls_open);
1373	while (nstp != LIST_END(&stp->ls_open)) {
1374		tstp = nstp;
1375		nstp = LIST_NEXT(nstp, ls_list);
1376		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1377	}
1378	if (stp->ls_op)
1379		nfsrvd_derefcache(stp->ls_op);
1380	FREE((caddr_t)stp, M_NFSDSTATE);
1381	newnfsstats.srvopenowners--;
1382	nfsrv_openpluslock--;
1383}
1384
1385/*
1386 * This function frees an open (nfsstate open structure) with all associated
1387 * lock_owners and locks. It also frees the nfslockfile structure iff there
1388 * are no other opens on the file.
1389 * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1390 */
1391static int
1392nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1393{
1394	struct nfsstate *nstp, *tstp;
1395	struct nfslockfile *lfp;
1396	int ret;
1397
1398	LIST_REMOVE(stp, ls_hash);
1399	LIST_REMOVE(stp, ls_list);
1400	LIST_REMOVE(stp, ls_file);
1401
1402	lfp = stp->ls_lfp;
1403	/*
1404	 * Now, free all lockowners associated with this open.
1405	 */
1406	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1407		nfsrv_freelockowner(tstp, vp, cansleep, p);
1408
1409	/*
1410	 * The nfslockfile is freed here if there are no locks
1411	 * associated with the open.
1412	 * If there are locks associated with the open, the
1413	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1414	 * Acquire the state mutex to avoid races with calls to
1415	 * nfsrv_getlockfile().
1416	 */
1417	if (cansleep != 0)
1418		NFSLOCKSTATE();
1419	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1420	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1421	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1422	    lfp->lf_usecount == 0 &&
1423	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1424		nfsrv_freenfslockfile(lfp);
1425		ret = 1;
1426	} else
1427		ret = 0;
1428	if (cansleep != 0)
1429		NFSUNLOCKSTATE();
1430	FREE((caddr_t)stp, M_NFSDSTATE);
1431	newnfsstats.srvopens--;
1432	nfsrv_openpluslock--;
1433	return (ret);
1434}
1435
1436/*
1437 * Frees a lockowner and all associated locks.
1438 */
1439static void
1440nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1441    NFSPROC_T *p)
1442{
1443
1444	LIST_REMOVE(stp, ls_hash);
1445	LIST_REMOVE(stp, ls_list);
1446	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1447	if (stp->ls_op)
1448		nfsrvd_derefcache(stp->ls_op);
1449	FREE((caddr_t)stp, M_NFSDSTATE);
1450	newnfsstats.srvlockowners--;
1451	nfsrv_openpluslock--;
1452}
1453
1454/*
1455 * Free all the nfs locks on a lockowner.
1456 */
1457static void
1458nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1459    NFSPROC_T *p)
1460{
1461	struct nfslock *lop, *nlop;
1462	struct nfsrollback *rlp, *nrlp;
1463	struct nfslockfile *lfp = NULL;
1464	int gottvp = 0;
1465	vnode_t tvp = NULL;
1466	uint64_t first, end;
1467
1468	if (vp != NULL)
1469		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1470	lop = LIST_FIRST(&stp->ls_lock);
1471	while (lop != LIST_END(&stp->ls_lock)) {
1472		nlop = LIST_NEXT(lop, lo_lckowner);
1473		/*
1474		 * Since all locks should be for the same file, lfp should
1475		 * not change.
1476		 */
1477		if (lfp == NULL)
1478			lfp = lop->lo_lfp;
1479		else if (lfp != lop->lo_lfp)
1480			panic("allnfslocks");
1481		/*
1482		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1483		 * from the file handle. This only occurs when called from
1484		 * nfsrv_cleanclient().
1485		 */
1486		if (gottvp == 0) {
1487			if (nfsrv_dolocallocks == 0)
1488				tvp = NULL;
1489			else if (vp == NULL && cansleep != 0) {
1490				tvp = nfsvno_getvp(&lfp->lf_fh);
1491				if (tvp != NULL)
1492					NFSVOPUNLOCK(tvp, 0);
1493			} else
1494				tvp = vp;
1495			gottvp = 1;
1496		}
1497
1498		if (tvp != NULL) {
1499			if (cansleep == 0)
1500				panic("allnfs2");
1501			first = lop->lo_first;
1502			end = lop->lo_end;
1503			nfsrv_freenfslock(lop);
1504			nfsrv_localunlock(tvp, lfp, first, end, p);
1505			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1506			    nrlp)
1507				free(rlp, M_NFSDROLLBACK);
1508			LIST_INIT(&lfp->lf_rollback);
1509		} else
1510			nfsrv_freenfslock(lop);
1511		lop = nlop;
1512	}
1513	if (vp == NULL && tvp != NULL)
1514		vrele(tvp);
1515}
1516
1517/*
1518 * Free an nfslock structure.
1519 */
1520static void
1521nfsrv_freenfslock(struct nfslock *lop)
1522{
1523
1524	if (lop->lo_lckfile.le_prev != NULL) {
1525		LIST_REMOVE(lop, lo_lckfile);
1526		newnfsstats.srvlocks--;
1527		nfsrv_openpluslock--;
1528	}
1529	LIST_REMOVE(lop, lo_lckowner);
1530	FREE((caddr_t)lop, M_NFSDLOCK);
1531}
1532
1533/*
1534 * This function frees an nfslockfile structure.
1535 */
1536static void
1537nfsrv_freenfslockfile(struct nfslockfile *lfp)
1538{
1539
1540	LIST_REMOVE(lfp, lf_hash);
1541	FREE((caddr_t)lfp, M_NFSDLOCKFILE);
1542}
1543
1544/*
1545 * This function looks up an nfsstate structure via stateid.
1546 */
1547static int
1548nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1549    struct nfsstate **stpp)
1550{
1551	struct nfsstate *stp;
1552	struct nfsstatehead *hp;
1553	int error = 0;
1554
1555	*stpp = NULL;
1556	hp = NFSSTATEHASH(clp, *stateidp);
1557	LIST_FOREACH(stp, hp, ls_hash) {
1558		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1559			NFSX_STATEIDOTHER))
1560			break;
1561	}
1562
1563	/*
1564	 * If no state id in list, return NFSERR_BADSTATEID.
1565	 */
1566	if (stp == LIST_END(hp)) {
1567		error = NFSERR_BADSTATEID;
1568		goto out;
1569	}
1570	*stpp = stp;
1571
1572out:
1573	NFSEXITCODE(error);
1574	return (error);
1575}
1576
1577/*
1578 * This function gets an nfsstate structure via owner string.
1579 */
1580static void
1581nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1582    struct nfsstate **stpp)
1583{
1584	struct nfsstate *stp;
1585
1586	*stpp = NULL;
1587	LIST_FOREACH(stp, hp, ls_list) {
1588		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1589		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1590			*stpp = stp;
1591			return;
1592		}
1593	}
1594}
1595
1596/*
1597 * Lock control function called to update lock status.
1598 * Returns 0 upon success, -1 if there is no lock and the flags indicate
1599 * that one isn't to be created and an NFSERR_xxx for other errors.
1600 * The structures new_stp and new_lop are passed in as pointers that should
1601 * be set to NULL if the structure is used and shouldn't be free'd.
1602 * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1603 * never used and can safely be allocated on the stack. For all other
1604 * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1605 * in case they are used.
1606 */
1607APPLESTATIC int
1608nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1609    struct nfslock **new_lopp, struct nfslockconflict *cfp,
1610    nfsquad_t clientid, nfsv4stateid_t *stateidp,
1611    __unused struct nfsexstuff *exp,
1612    struct nfsrv_descript *nd, NFSPROC_T *p)
1613{
1614	struct nfslock *lop;
1615	struct nfsstate *new_stp = *new_stpp;
1616	struct nfslock *new_lop = *new_lopp;
1617	struct nfsstate *tstp, *mystp, *nstp;
1618	int specialid = 0;
1619	struct nfslockfile *lfp;
1620	struct nfslock *other_lop = NULL;
1621	struct nfsstate *stp, *lckstp = NULL;
1622	struct nfsclient *clp = NULL;
1623	u_int32_t bits;
1624	int error = 0, haslock = 0, ret, reterr;
1625	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1626	fhandle_t nfh;
1627	uint64_t first, end;
1628	uint32_t lock_flags;
1629
1630	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1631		/*
1632		 * Note the special cases of "all 1s" or "all 0s" stateids and
1633		 * let reads with all 1s go ahead.
1634		 */
1635		if (new_stp->ls_stateid.seqid == 0x0 &&
1636		    new_stp->ls_stateid.other[0] == 0x0 &&
1637		    new_stp->ls_stateid.other[1] == 0x0 &&
1638		    new_stp->ls_stateid.other[2] == 0x0)
1639			specialid = 1;
1640		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1641		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1642		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1643		    new_stp->ls_stateid.other[2] == 0xffffffff)
1644			specialid = 2;
1645	}
1646
1647	/*
1648	 * Check for restart conditions (client and server).
1649	 */
1650	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1651	    &new_stp->ls_stateid, specialid);
1652	if (error)
1653		goto out;
1654
1655	/*
1656	 * Check for state resource limit exceeded.
1657	 */
1658	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1659	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1660		error = NFSERR_RESOURCE;
1661		goto out;
1662	}
1663
1664	/*
1665	 * For the lock case, get another nfslock structure,
1666	 * just in case we need it.
1667	 * Malloc now, before we start sifting through the linked lists,
1668	 * in case we have to wait for memory.
1669	 */
1670tryagain:
1671	if (new_stp->ls_flags & NFSLCK_LOCK)
1672		MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
1673		    M_NFSDLOCK, M_WAITOK);
1674	filestruct_locked = 0;
1675	reterr = 0;
1676	lfp = NULL;
1677
1678	/*
1679	 * Get the lockfile structure for CFH now, so we can do a sanity
1680	 * check against the stateid, before incrementing the seqid#, since
1681	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1682	 * shouldn't be incremented for this case.
1683	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1684	 * will be handled later.
1685	 * If we are doing Lock/LockU and local locking is enabled, sleep
1686	 * lock the nfslockfile structure.
1687	 */
1688	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1689	NFSLOCKSTATE();
1690	if (getlckret == 0) {
1691		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1692		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1693			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1694			    &lfp, &nfh, 1);
1695			if (getlckret == 0)
1696				filestruct_locked = 1;
1697		} else
1698			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1699			    &lfp, &nfh, 0);
1700	}
1701	if (getlckret != 0 && getlckret != -1)
1702		reterr = getlckret;
1703
1704	if (filestruct_locked != 0) {
1705		LIST_INIT(&lfp->lf_rollback);
1706		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1707			/*
1708			 * For local locking, do the advisory locking now, so
1709			 * that any conflict can be detected. A failure later
1710			 * can be rolled back locally. If an error is returned,
1711			 * struct nfslockfile has been unlocked and any local
1712			 * locking rolled back.
1713			 */
1714			NFSUNLOCKSTATE();
1715			if (vnode_unlocked == 0) {
1716				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1717				vnode_unlocked = 1;
1718				NFSVOPUNLOCK(vp, 0);
1719			}
1720			reterr = nfsrv_locallock(vp, lfp,
1721			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1722			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1723			NFSLOCKSTATE();
1724		}
1725	}
1726
1727	if (specialid == 0) {
1728	    if (new_stp->ls_flags & NFSLCK_TEST) {
1729		/*
1730		 * RFC 3530 does not list LockT as an op that renews a
1731		 * lease, but the concensus seems to be that it is ok
1732		 * for a server to do so.
1733		 */
1734		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1735		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1736
1737		/*
1738		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1739		 * error returns for LockT, just go ahead and test for a lock,
1740		 * since there are no locks for this client, but other locks
1741		 * can conflict. (ie. same client will always be false)
1742		 */
1743		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1744		    error = 0;
1745		lckstp = new_stp;
1746	    } else {
1747	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1748		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1749	      if (error == 0)
1750		/*
1751		 * Look up the stateid
1752		 */
1753		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1754		  new_stp->ls_flags, &stp);
1755	      /*
1756	       * do some sanity checks for an unconfirmed open or a
1757	       * stateid that refers to the wrong file, for an open stateid
1758	       */
1759	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1760		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1761		   (getlckret == 0 && stp->ls_lfp != lfp)))
1762			error = NFSERR_BADSTATEID;
1763	      if (error == 0 &&
1764		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1765		  getlckret == 0 && stp->ls_lfp != lfp)
1766			error = NFSERR_BADSTATEID;
1767
1768	      /*
1769	       * If the lockowner stateid doesn't refer to the same file,
1770	       * I believe that is considered ok, since some clients will
1771	       * only create a single lockowner and use that for all locks
1772	       * on all files.
1773	       * For now, log it as a diagnostic, instead of considering it
1774	       * a BadStateid.
1775	       */
1776	      if (error == 0 && (stp->ls_flags &
1777		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1778		  getlckret == 0 && stp->ls_lfp != lfp) {
1779#ifdef DIAGNOSTIC
1780		  printf("Got a lock statid for different file open\n");
1781#endif
1782		  /*
1783		  error = NFSERR_BADSTATEID;
1784		  */
1785	      }
1786
1787	      if (error == 0) {
1788		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1789			/*
1790			 * If haslock set, we've already checked the seqid.
1791			 */
1792			if (!haslock) {
1793			    if (stp->ls_flags & NFSLCK_OPEN)
1794				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1795				    stp->ls_openowner, new_stp->ls_op);
1796			    else
1797				error = NFSERR_BADSTATEID;
1798			}
1799			if (!error)
1800			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1801			if (lckstp)
1802			    /*
1803			     * I believe this should be an error, but it
1804			     * isn't obvious what NFSERR_xxx would be
1805			     * appropriate, so I'll use NFSERR_INVAL for now.
1806			     */
1807			    error = NFSERR_INVAL;
1808			else
1809			    lckstp = new_stp;
1810		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1811			/*
1812			 * If haslock set, ditto above.
1813			 */
1814			if (!haslock) {
1815			    if (stp->ls_flags & NFSLCK_OPEN)
1816				error = NFSERR_BADSTATEID;
1817			    else
1818				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1819				    stp, new_stp->ls_op);
1820			}
1821			lckstp = stp;
1822		    } else {
1823			lckstp = stp;
1824		    }
1825	      }
1826	      /*
1827	       * If the seqid part of the stateid isn't the same, return
1828	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1829	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1830	       * nfsrv_returnoldstateid is set. (The concensus on the email
1831	       * list was that most clients would prefer to not receive
1832	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1833	       * is what will happen, so I use the nfsrv_returnoldstateid to
1834	       * allow for either server configuration.)
1835	       */
1836	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1837		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1838		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1839		    nfsrv_returnoldstateid)) ||
1840		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1841		    new_stp->ls_stateid.seqid != 0)))
1842		    error = NFSERR_OLDSTATEID;
1843	    }
1844	}
1845
1846	/*
1847	 * Now we can check for grace.
1848	 */
1849	if (!error)
1850		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1851	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1852		nfsrv_checkstable(clp))
1853		error = NFSERR_NOGRACE;
1854	/*
1855	 * If we successfully Reclaimed state, note that.
1856	 */
1857	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1858		nfsrv_markstable(clp);
1859
1860	/*
1861	 * At this point, either error == NFSERR_BADSTATEID or the
1862	 * seqid# has been updated, so we can return any error.
1863	 * If error == 0, there may be an error in:
1864	 *    nd_repstat - Set by the calling function.
1865	 *    reterr - Set above, if getting the nfslockfile structure
1866	 *       or acquiring the local lock failed.
1867	 *    (If both of these are set, nd_repstat should probably be
1868	 *     returned, since that error was detected before this
1869	 *     function call.)
1870	 */
1871	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1872		if (error == 0) {
1873			if (nd->nd_repstat != 0)
1874				error = nd->nd_repstat;
1875			else
1876				error = reterr;
1877		}
1878		if (filestruct_locked != 0) {
1879			/* Roll back local locks. */
1880			NFSUNLOCKSTATE();
1881			if (vnode_unlocked == 0) {
1882				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1883				vnode_unlocked = 1;
1884				NFSVOPUNLOCK(vp, 0);
1885			}
1886			nfsrv_locallock_rollback(vp, lfp, p);
1887			NFSLOCKSTATE();
1888			nfsrv_unlocklf(lfp);
1889		}
1890		NFSUNLOCKSTATE();
1891		goto out;
1892	}
1893
1894	/*
1895	 * Check the nfsrv_getlockfile return.
1896	 * Returned -1 if no structure found.
1897	 */
1898	if (getlckret == -1) {
1899		error = NFSERR_EXPIRED;
1900		/*
1901		 * Called from lockt, so no lock is OK.
1902		 */
1903		if (new_stp->ls_flags & NFSLCK_TEST) {
1904			error = 0;
1905		} else if (new_stp->ls_flags &
1906		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1907			/*
1908			 * Called to check for a lock, OK if the stateid is all
1909			 * 1s or all 0s, but there should be an nfsstate
1910			 * otherwise.
1911			 * (ie. If there is no open, I'll assume no share
1912			 *  deny bits.)
1913			 */
1914			if (specialid)
1915				error = 0;
1916			else
1917				error = NFSERR_BADSTATEID;
1918		}
1919		NFSUNLOCKSTATE();
1920		goto out;
1921	}
1922
1923	/*
1924	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1925	 * For NFSLCK_CHECK, allow a read if write access is granted,
1926	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1927	 * which implies a conflicting deny can't exist.
1928	 */
1929	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1930	    /*
1931	     * Four kinds of state id:
1932	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1933	     * - stateid for an open
1934	     * - stateid for a delegation
1935	     * - stateid for a lock owner
1936	     */
1937	    if (!specialid) {
1938		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1939		    delegation = 1;
1940		    mystp = stp;
1941		    nfsrv_delaydelegtimeout(stp);
1942	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1943		    mystp = stp;
1944		} else {
1945		    mystp = stp->ls_openstp;
1946		}
1947		/*
1948		 * If locking or checking, require correct access
1949		 * bit set.
1950		 */
1951		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1952		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1953		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1954		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1955		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1956		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
1957		     nfsrv_allowreadforwriteopen == 0) ||
1958		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1959		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1960		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1961			if (filestruct_locked != 0) {
1962				/* Roll back local locks. */
1963				NFSUNLOCKSTATE();
1964				if (vnode_unlocked == 0) {
1965					ASSERT_VOP_ELOCKED(vp,
1966					    "nfsrv_lockctrl3");
1967					vnode_unlocked = 1;
1968					NFSVOPUNLOCK(vp, 0);
1969				}
1970				nfsrv_locallock_rollback(vp, lfp, p);
1971				NFSLOCKSTATE();
1972				nfsrv_unlocklf(lfp);
1973			}
1974			NFSUNLOCKSTATE();
1975			error = NFSERR_OPENMODE;
1976			goto out;
1977		}
1978	    } else
1979		mystp = NULL;
1980	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1981		/*
1982		 * Check for a conflicting deny bit.
1983		 */
1984		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1985		    if (tstp != mystp) {
1986			bits = tstp->ls_flags;
1987			bits >>= NFSLCK_SHIFT;
1988			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1989			    KASSERT(vnode_unlocked == 0,
1990				("nfsrv_lockctrl: vnode unlocked1"));
1991			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1992				vp, p);
1993			    if (ret == 1) {
1994				/*
1995				* nfsrv_clientconflict unlocks state
1996				 * when it returns non-zero.
1997				 */
1998				lckstp = NULL;
1999				goto tryagain;
2000			    }
2001			    if (ret == 0)
2002				NFSUNLOCKSTATE();
2003			    if (ret == 2)
2004				error = NFSERR_PERM;
2005			    else
2006				error = NFSERR_OPENMODE;
2007			    goto out;
2008			}
2009		    }
2010		}
2011
2012		/* We're outta here */
2013		NFSUNLOCKSTATE();
2014		goto out;
2015	    }
2016	}
2017
2018	/*
2019	 * For setattr, just get rid of all the Delegations for other clients.
2020	 */
2021	if (new_stp->ls_flags & NFSLCK_SETATTR) {
2022		KASSERT(vnode_unlocked == 0,
2023		    ("nfsrv_lockctrl: vnode unlocked2"));
2024		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2025		if (ret) {
2026			/*
2027			 * nfsrv_cleandeleg() unlocks state when it
2028			 * returns non-zero.
2029			 */
2030			if (ret == -1) {
2031				lckstp = NULL;
2032				goto tryagain;
2033			}
2034			error = ret;
2035			goto out;
2036		}
2037		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2038		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2039		     LIST_EMPTY(&lfp->lf_deleg))) {
2040			NFSUNLOCKSTATE();
2041			goto out;
2042		}
2043	}
2044
2045	/*
2046	 * Check for a conflicting delegation. If one is found, call
2047	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2048	 * been set yet, it will get the lock. Otherwise, it will recall
2049	 * the delegation. Then, we try try again...
2050	 * I currently believe the conflict algorithm to be:
2051	 * For Lock Ops (Lock/LockT/LockU)
2052	 * - there is a conflict iff a different client has a write delegation
2053	 * For Reading (Read Op)
2054	 * - there is a conflict iff a different client has a write delegation
2055	 *   (the specialids are always a different client)
2056	 * For Writing (Write/Setattr of size)
2057	 * - there is a conflict if a different client has any delegation
2058	 * - there is a conflict if the same client has a read delegation
2059	 *   (I don't understand why this isn't allowed, but that seems to be
2060	 *    the current concensus?)
2061	 */
2062	tstp = LIST_FIRST(&lfp->lf_deleg);
2063	while (tstp != LIST_END(&lfp->lf_deleg)) {
2064	    nstp = LIST_NEXT(tstp, ls_file);
2065	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2066		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2067		  (new_lop->lo_flags & NFSLCK_READ))) &&
2068		  clp != tstp->ls_clp &&
2069		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2070		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2071		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2072		  (clp != tstp->ls_clp ||
2073		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2074		ret = 0;
2075		if (filestruct_locked != 0) {
2076			/* Roll back local locks. */
2077			NFSUNLOCKSTATE();
2078			if (vnode_unlocked == 0) {
2079				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2080				NFSVOPUNLOCK(vp, 0);
2081			}
2082			nfsrv_locallock_rollback(vp, lfp, p);
2083			NFSLOCKSTATE();
2084			nfsrv_unlocklf(lfp);
2085			NFSUNLOCKSTATE();
2086			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2087			vnode_unlocked = 0;
2088			if ((vp->v_iflag & VI_DOOMED) != 0)
2089				ret = NFSERR_SERVERFAULT;
2090			NFSLOCKSTATE();
2091		}
2092		if (ret == 0)
2093			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2094		if (ret) {
2095		    /*
2096		     * nfsrv_delegconflict unlocks state when it
2097		     * returns non-zero, which it always does.
2098		     */
2099		    if (other_lop) {
2100			FREE((caddr_t)other_lop, M_NFSDLOCK);
2101			other_lop = NULL;
2102		    }
2103		    if (ret == -1) {
2104			lckstp = NULL;
2105			goto tryagain;
2106		    }
2107		    error = ret;
2108		    goto out;
2109		}
2110		/* Never gets here. */
2111	    }
2112	    tstp = nstp;
2113	}
2114
2115	/*
2116	 * Handle the unlock case by calling nfsrv_updatelock().
2117	 * (Should I have done some access checking above for unlock? For now,
2118	 *  just let it happen.)
2119	 */
2120	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2121		first = new_lop->lo_first;
2122		end = new_lop->lo_end;
2123		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2124		stateidp->seqid = ++(stp->ls_stateid.seqid);
2125		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2126			stateidp->seqid = stp->ls_stateid.seqid = 1;
2127		stateidp->other[0] = stp->ls_stateid.other[0];
2128		stateidp->other[1] = stp->ls_stateid.other[1];
2129		stateidp->other[2] = stp->ls_stateid.other[2];
2130		if (filestruct_locked != 0) {
2131			NFSUNLOCKSTATE();
2132			if (vnode_unlocked == 0) {
2133				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2134				vnode_unlocked = 1;
2135				NFSVOPUNLOCK(vp, 0);
2136			}
2137			/* Update the local locks. */
2138			nfsrv_localunlock(vp, lfp, first, end, p);
2139			NFSLOCKSTATE();
2140			nfsrv_unlocklf(lfp);
2141		}
2142		NFSUNLOCKSTATE();
2143		goto out;
2144	}
2145
2146	/*
2147	 * Search for a conflicting lock. A lock conflicts if:
2148	 * - the lock range overlaps and
2149	 * - at least one lock is a write lock and
2150	 * - it is not owned by the same lock owner
2151	 */
2152	if (!delegation) {
2153	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2154	    if (new_lop->lo_end > lop->lo_first &&
2155		new_lop->lo_first < lop->lo_end &&
2156		(new_lop->lo_flags == NFSLCK_WRITE ||
2157		 lop->lo_flags == NFSLCK_WRITE) &&
2158		lckstp != lop->lo_stp &&
2159		(clp != lop->lo_stp->ls_clp ||
2160		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2161		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2162		    lckstp->ls_ownerlen))) {
2163		if (other_lop) {
2164		    FREE((caddr_t)other_lop, M_NFSDLOCK);
2165		    other_lop = NULL;
2166		}
2167		if (vnode_unlocked != 0)
2168		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2169			NULL, p);
2170		else
2171		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2172			vp, p);
2173		if (ret == 1) {
2174		    if (filestruct_locked != 0) {
2175			if (vnode_unlocked == 0) {
2176				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2177				NFSVOPUNLOCK(vp, 0);
2178			}
2179			/* Roll back local locks. */
2180			nfsrv_locallock_rollback(vp, lfp, p);
2181			NFSLOCKSTATE();
2182			nfsrv_unlocklf(lfp);
2183			NFSUNLOCKSTATE();
2184			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2185			vnode_unlocked = 0;
2186			if ((vp->v_iflag & VI_DOOMED) != 0) {
2187				error = NFSERR_SERVERFAULT;
2188				goto out;
2189			}
2190		    }
2191		    /*
2192		     * nfsrv_clientconflict() unlocks state when it
2193		     * returns non-zero.
2194		     */
2195		    lckstp = NULL;
2196		    goto tryagain;
2197		}
2198		/*
2199		 * Found a conflicting lock, so record the conflict and
2200		 * return the error.
2201		 */
2202		if (cfp != NULL && ret == 0) {
2203		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2204		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2205		    cfp->cl_first = lop->lo_first;
2206		    cfp->cl_end = lop->lo_end;
2207		    cfp->cl_flags = lop->lo_flags;
2208		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2209		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2210			cfp->cl_ownerlen);
2211		}
2212		if (ret == 2)
2213		    error = NFSERR_PERM;
2214		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2215		    error = NFSERR_RECLAIMCONFLICT;
2216		else if (new_stp->ls_flags & NFSLCK_CHECK)
2217		    error = NFSERR_LOCKED;
2218		else
2219		    error = NFSERR_DENIED;
2220		if (filestruct_locked != 0 && ret == 0) {
2221			/* Roll back local locks. */
2222			NFSUNLOCKSTATE();
2223			if (vnode_unlocked == 0) {
2224				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2225				vnode_unlocked = 1;
2226				NFSVOPUNLOCK(vp, 0);
2227			}
2228			nfsrv_locallock_rollback(vp, lfp, p);
2229			NFSLOCKSTATE();
2230			nfsrv_unlocklf(lfp);
2231		}
2232		if (ret == 0)
2233			NFSUNLOCKSTATE();
2234		goto out;
2235	    }
2236	  }
2237	}
2238
2239	/*
2240	 * We only get here if there was no lock that conflicted.
2241	 */
2242	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2243		NFSUNLOCKSTATE();
2244		goto out;
2245	}
2246
2247	/*
2248	 * We only get here when we are creating or modifying a lock.
2249	 * There are two variants:
2250	 * - exist_lock_owner where lock_owner exists
2251	 * - open_to_lock_owner with new lock_owner
2252	 */
2253	first = new_lop->lo_first;
2254	end = new_lop->lo_end;
2255	lock_flags = new_lop->lo_flags;
2256	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2257		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2258		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2259		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2260			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2261		stateidp->other[0] = lckstp->ls_stateid.other[0];
2262		stateidp->other[1] = lckstp->ls_stateid.other[1];
2263		stateidp->other[2] = lckstp->ls_stateid.other[2];
2264	} else {
2265		/*
2266		 * The new open_to_lock_owner case.
2267		 * Link the new nfsstate into the lists.
2268		 */
2269		new_stp->ls_seq = new_stp->ls_opentolockseq;
2270		nfsrvd_refcache(new_stp->ls_op);
2271		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2272		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2273		    clp->lc_clientid.lval[0];
2274		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2275		    clp->lc_clientid.lval[1];
2276		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2277		    nfsrv_nextstateindex(clp);
2278		new_stp->ls_clp = clp;
2279		LIST_INIT(&new_stp->ls_lock);
2280		new_stp->ls_openstp = stp;
2281		new_stp->ls_lfp = lfp;
2282		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2283		    lfp);
2284		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2285		    new_stp, ls_hash);
2286		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2287		*new_lopp = NULL;
2288		*new_stpp = NULL;
2289		newnfsstats.srvlockowners++;
2290		nfsrv_openpluslock++;
2291	}
2292	if (filestruct_locked != 0) {
2293		NFSUNLOCKSTATE();
2294		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2295		NFSLOCKSTATE();
2296		nfsrv_unlocklf(lfp);
2297	}
2298	NFSUNLOCKSTATE();
2299
2300out:
2301	if (haslock) {
2302		NFSLOCKV4ROOTMUTEX();
2303		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2304		NFSUNLOCKV4ROOTMUTEX();
2305	}
2306	if (vnode_unlocked != 0) {
2307		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2308		if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0)
2309			error = NFSERR_SERVERFAULT;
2310	}
2311	if (other_lop)
2312		FREE((caddr_t)other_lop, M_NFSDLOCK);
2313	NFSEXITCODE2(error, nd);
2314	return (error);
2315}
2316
2317/*
2318 * Check for state errors for Open.
2319 * repstat is passed back out as an error if more critical errors
2320 * are not detected.
2321 */
2322APPLESTATIC int
2323nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2324    struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2325    NFSPROC_T *p, int repstat)
2326{
2327	struct nfsstate *stp, *nstp;
2328	struct nfsclient *clp;
2329	struct nfsstate *ownerstp;
2330	struct nfslockfile *lfp, *new_lfp;
2331	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2332
2333	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2334		readonly = 1;
2335	/*
2336	 * Check for restart conditions (client and server).
2337	 */
2338	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2339		&new_stp->ls_stateid, 0);
2340	if (error)
2341		goto out;
2342
2343	/*
2344	 * Check for state resource limit exceeded.
2345	 * Technically this should be SMP protected, but the worst
2346	 * case error is "out by one or two" on the count when it
2347	 * returns NFSERR_RESOURCE and the limit is just a rather
2348	 * arbitrary high water mark, so no harm is done.
2349	 */
2350	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2351		error = NFSERR_RESOURCE;
2352		goto out;
2353	}
2354
2355tryagain:
2356	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2357	    M_NFSDLOCKFILE, M_WAITOK);
2358	if (vp)
2359		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2360		    NULL, p);
2361	NFSLOCKSTATE();
2362	/*
2363	 * Get the nfsclient structure.
2364	 */
2365	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2366	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2367
2368	/*
2369	 * Look up the open owner. See if it needs confirmation and
2370	 * check the seq#, as required.
2371	 */
2372	if (!error)
2373		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2374
2375	if (!error && ownerstp) {
2376		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2377		    new_stp->ls_op);
2378		/*
2379		 * If the OpenOwner hasn't been confirmed, assume the
2380		 * old one was a replay and this one is ok.
2381		 * See: RFC3530 Sec. 14.2.18.
2382		 */
2383		if (error == NFSERR_BADSEQID &&
2384		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2385			error = 0;
2386	}
2387
2388	/*
2389	 * Check for grace.
2390	 */
2391	if (!error)
2392		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2393	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2394		nfsrv_checkstable(clp))
2395		error = NFSERR_NOGRACE;
2396
2397	/*
2398	 * If none of the above errors occurred, let repstat be
2399	 * returned.
2400	 */
2401	if (repstat && !error)
2402		error = repstat;
2403	if (error) {
2404		NFSUNLOCKSTATE();
2405		if (haslock) {
2406			NFSLOCKV4ROOTMUTEX();
2407			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2408			NFSUNLOCKV4ROOTMUTEX();
2409		}
2410		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2411		goto out;
2412	}
2413
2414	/*
2415	 * If vp == NULL, the file doesn't exist yet, so return ok.
2416	 * (This always happens on the first pass, so haslock must be 0.)
2417	 */
2418	if (vp == NULL) {
2419		NFSUNLOCKSTATE();
2420		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2421		goto out;
2422	}
2423
2424	/*
2425	 * Get the structure for the underlying file.
2426	 */
2427	if (getfhret)
2428		error = getfhret;
2429	else
2430		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2431		    NULL, 0);
2432	if (new_lfp)
2433		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2434	if (error) {
2435		NFSUNLOCKSTATE();
2436		if (haslock) {
2437			NFSLOCKV4ROOTMUTEX();
2438			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2439			NFSUNLOCKV4ROOTMUTEX();
2440		}
2441		goto out;
2442	}
2443
2444	/*
2445	 * Search for a conflicting open/share.
2446	 */
2447	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2448	    /*
2449	     * For Delegate_Cur, search for the matching Delegation,
2450	     * which indicates no conflict.
2451	     * An old delegation should have been recovered by the
2452	     * client doing a Claim_DELEGATE_Prev, so I won't let
2453	     * it match and return NFSERR_EXPIRED. Should I let it
2454	     * match?
2455	     */
2456	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2457		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2458		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2459		    stateidp->seqid == 0) ||
2460		    stateidp->seqid == stp->ls_stateid.seqid) &&
2461		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2462			  NFSX_STATEIDOTHER))
2463			break;
2464	    }
2465	    if (stp == LIST_END(&lfp->lf_deleg) ||
2466		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2467		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2468		NFSUNLOCKSTATE();
2469		if (haslock) {
2470			NFSLOCKV4ROOTMUTEX();
2471			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2472			NFSUNLOCKV4ROOTMUTEX();
2473		}
2474		error = NFSERR_EXPIRED;
2475		goto out;
2476	    }
2477	}
2478
2479	/*
2480	 * Check for access/deny bit conflicts. I check for the same
2481	 * owner as well, in case the client didn't bother.
2482	 */
2483	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2484		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2485		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2486		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2487		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2488		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2489			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2490			if (ret == 1) {
2491				/*
2492				 * nfsrv_clientconflict() unlocks
2493				 * state when it returns non-zero.
2494				 */
2495				goto tryagain;
2496			}
2497			if (ret == 2)
2498				error = NFSERR_PERM;
2499			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2500				error = NFSERR_RECLAIMCONFLICT;
2501			else
2502				error = NFSERR_SHAREDENIED;
2503			if (ret == 0)
2504				NFSUNLOCKSTATE();
2505			if (haslock) {
2506				NFSLOCKV4ROOTMUTEX();
2507				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2508				NFSUNLOCKV4ROOTMUTEX();
2509			}
2510			goto out;
2511		}
2512	}
2513
2514	/*
2515	 * Check for a conflicting delegation. If one is found, call
2516	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2517	 * been set yet, it will get the lock. Otherwise, it will recall
2518	 * the delegation. Then, we try try again...
2519	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2520	 *  isn't a conflict.)
2521	 * I currently believe the conflict algorithm to be:
2522	 * For Open with Read Access and Deny None
2523	 * - there is a conflict iff a different client has a write delegation
2524	 * For Open with other Write Access or any Deny except None
2525	 * - there is a conflict if a different client has any delegation
2526	 * - there is a conflict if the same client has a read delegation
2527	 *   (The current concensus is that this last case should be
2528	 *    considered a conflict since the client with a read delegation
2529	 *    could have done an Open with ReadAccess and WriteDeny
2530	 *    locally and then not have checked for the WriteDeny.)
2531	 * Don't check for a Reclaim, since that will be dealt with
2532	 * by nfsrv_openctrl().
2533	 */
2534	if (!(new_stp->ls_flags &
2535		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2536	    stp = LIST_FIRST(&lfp->lf_deleg);
2537	    while (stp != LIST_END(&lfp->lf_deleg)) {
2538		nstp = LIST_NEXT(stp, ls_file);
2539		if ((readonly && stp->ls_clp != clp &&
2540		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2541		    (!readonly && (stp->ls_clp != clp ||
2542		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2543			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2544			if (ret) {
2545			    /*
2546			     * nfsrv_delegconflict() unlocks state
2547			     * when it returns non-zero.
2548			     */
2549			    if (ret == -1)
2550				goto tryagain;
2551			    error = ret;
2552			    goto out;
2553			}
2554		}
2555		stp = nstp;
2556	    }
2557	}
2558	NFSUNLOCKSTATE();
2559	if (haslock) {
2560		NFSLOCKV4ROOTMUTEX();
2561		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2562		NFSUNLOCKV4ROOTMUTEX();
2563	}
2564
2565out:
2566	NFSEXITCODE2(error, nd);
2567	return (error);
2568}
2569
2570/*
2571 * Open control function to create/update open state for an open.
2572 */
2573APPLESTATIC int
2574nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2575    struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2576    nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2577    NFSPROC_T *p, u_quad_t filerev)
2578{
2579	struct nfsstate *new_stp = *new_stpp;
2580	struct nfsstate *stp, *nstp;
2581	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2582	struct nfslockfile *lfp, *new_lfp;
2583	struct nfsclient *clp;
2584	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2585	int readonly = 0, cbret = 1, getfhret = 0;
2586	int gotstate = 0, len = 0;
2587	u_char *clidp = NULL;
2588
2589	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2590		readonly = 1;
2591	/*
2592	 * Check for restart conditions (client and server).
2593	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2594	 * If an error does show up, return NFSERR_EXPIRED, since the
2595	 * the seqid# has already been incremented.
2596	 */
2597	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2598	    &new_stp->ls_stateid, 0);
2599	if (error) {
2600		printf("Nfsd: openctrl unexpected restart err=%d\n",
2601		    error);
2602		error = NFSERR_EXPIRED;
2603		goto out;
2604	}
2605
2606	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2607tryagain:
2608	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2609	    M_NFSDLOCKFILE, M_WAITOK);
2610	MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2611	    M_NFSDSTATE, M_WAITOK);
2612	MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2613	    M_NFSDSTATE, M_WAITOK);
2614	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2615	    NULL, p);
2616	NFSLOCKSTATE();
2617	/*
2618	 * Get the client structure. Since the linked lists could be changed
2619	 * by other nfsd processes if this process does a tsleep(), one of
2620	 * two things must be done.
2621	 * 1 - don't tsleep()
2622	 * or
2623	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2624	 *     before using the lists, since this lock stops the other
2625	 *     nfsd. This should only be used for rare cases, since it
2626	 *     essentially single threads the nfsd.
2627	 *     At this time, it is only done for cases where the stable
2628	 *     storage file must be written prior to completion of state
2629	 *     expiration.
2630	 */
2631	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2632	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2633	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2634	    clp->lc_program) {
2635		/*
2636		 * This happens on the first open for a client
2637		 * that supports callbacks.
2638		 */
2639		NFSUNLOCKSTATE();
2640		/*
2641		 * Although nfsrv_docallback() will sleep, clp won't
2642		 * go away, since they are only removed when the
2643		 * nfsv4_lock() has blocked the nfsd threads. The
2644		 * fields in clp can change, but having multiple
2645		 * threads do this Null callback RPC should be
2646		 * harmless.
2647		 */
2648		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2649		    NULL, 0, NULL, NULL, NULL, p);
2650		NFSLOCKSTATE();
2651		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2652		if (!cbret)
2653			clp->lc_flags |= LCL_CALLBACKSON;
2654	}
2655
2656	/*
2657	 * Look up the open owner. See if it needs confirmation and
2658	 * check the seq#, as required.
2659	 */
2660	if (!error)
2661		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2662
2663	if (error) {
2664		NFSUNLOCKSTATE();
2665		printf("Nfsd: openctrl unexpected state err=%d\n",
2666			error);
2667		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2668		free((caddr_t)new_open, M_NFSDSTATE);
2669		free((caddr_t)new_deleg, M_NFSDSTATE);
2670		if (haslock) {
2671			NFSLOCKV4ROOTMUTEX();
2672			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2673			NFSUNLOCKV4ROOTMUTEX();
2674		}
2675		error = NFSERR_EXPIRED;
2676		goto out;
2677	}
2678
2679	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2680		nfsrv_markstable(clp);
2681
2682	/*
2683	 * Get the structure for the underlying file.
2684	 */
2685	if (getfhret)
2686		error = getfhret;
2687	else
2688		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2689		    NULL, 0);
2690	if (new_lfp)
2691		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2692	if (error) {
2693		NFSUNLOCKSTATE();
2694		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2695		    error);
2696		free((caddr_t)new_open, M_NFSDSTATE);
2697		free((caddr_t)new_deleg, M_NFSDSTATE);
2698		if (haslock) {
2699			NFSLOCKV4ROOTMUTEX();
2700			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2701			NFSUNLOCKV4ROOTMUTEX();
2702		}
2703		goto out;
2704	}
2705
2706	/*
2707	 * Search for a conflicting open/share.
2708	 */
2709	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2710	    /*
2711	     * For Delegate_Cur, search for the matching Delegation,
2712	     * which indicates no conflict.
2713	     * An old delegation should have been recovered by the
2714	     * client doing a Claim_DELEGATE_Prev, so I won't let
2715	     * it match and return NFSERR_EXPIRED. Should I let it
2716	     * match?
2717	     */
2718	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2719		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2720		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2721		    stateidp->seqid == 0) ||
2722		    stateidp->seqid == stp->ls_stateid.seqid) &&
2723		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2724			NFSX_STATEIDOTHER))
2725			break;
2726	    }
2727	    if (stp == LIST_END(&lfp->lf_deleg) ||
2728		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2729		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2730		NFSUNLOCKSTATE();
2731		printf("Nfsd openctrl unexpected expiry\n");
2732		free((caddr_t)new_open, M_NFSDSTATE);
2733		free((caddr_t)new_deleg, M_NFSDSTATE);
2734		if (haslock) {
2735			NFSLOCKV4ROOTMUTEX();
2736			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2737			NFSUNLOCKV4ROOTMUTEX();
2738		}
2739		error = NFSERR_EXPIRED;
2740		goto out;
2741	    }
2742
2743	    /*
2744	     * Don't issue a Delegation, since one already exists and
2745	     * delay delegation timeout, as required.
2746	     */
2747	    delegate = 0;
2748	    nfsrv_delaydelegtimeout(stp);
2749	}
2750
2751	/*
2752	 * Check for access/deny bit conflicts. I also check for the
2753	 * same owner, since the client might not have bothered to check.
2754	 * Also, note an open for the same file and owner, if found,
2755	 * which is all we do here for Delegate_Cur, since conflict
2756	 * checking is already done.
2757	 */
2758	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2759		if (ownerstp && stp->ls_openowner == ownerstp)
2760			openstp = stp;
2761		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2762		    /*
2763		     * If another client has the file open, the only
2764		     * delegation that can be issued is a Read delegation
2765		     * and only if it is a Read open with Deny none.
2766		     */
2767		    if (clp != stp->ls_clp) {
2768			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2769			    NFSLCK_READACCESS)
2770			    writedeleg = 0;
2771			else
2772			    delegate = 0;
2773		    }
2774		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2775		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2776		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2777		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2778			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2779			if (ret == 1) {
2780				/*
2781				 * nfsrv_clientconflict() unlocks state
2782				 * when it returns non-zero.
2783				 */
2784				free((caddr_t)new_open, M_NFSDSTATE);
2785				free((caddr_t)new_deleg, M_NFSDSTATE);
2786				openstp = NULL;
2787				goto tryagain;
2788			}
2789			if (ret == 2)
2790				error = NFSERR_PERM;
2791			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2792				error = NFSERR_RECLAIMCONFLICT;
2793			else
2794				error = NFSERR_SHAREDENIED;
2795			if (ret == 0)
2796				NFSUNLOCKSTATE();
2797			if (haslock) {
2798				NFSLOCKV4ROOTMUTEX();
2799				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2800				NFSUNLOCKV4ROOTMUTEX();
2801			}
2802			free((caddr_t)new_open, M_NFSDSTATE);
2803			free((caddr_t)new_deleg, M_NFSDSTATE);
2804			printf("nfsd openctrl unexpected client cnfl\n");
2805			goto out;
2806		    }
2807		}
2808	}
2809
2810	/*
2811	 * Check for a conflicting delegation. If one is found, call
2812	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2813	 * been set yet, it will get the lock. Otherwise, it will recall
2814	 * the delegation. Then, we try try again...
2815	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2816	 *  isn't a conflict.)
2817	 * I currently believe the conflict algorithm to be:
2818	 * For Open with Read Access and Deny None
2819	 * - there is a conflict iff a different client has a write delegation
2820	 * For Open with other Write Access or any Deny except None
2821	 * - there is a conflict if a different client has any delegation
2822	 * - there is a conflict if the same client has a read delegation
2823	 *   (The current concensus is that this last case should be
2824	 *    considered a conflict since the client with a read delegation
2825	 *    could have done an Open with ReadAccess and WriteDeny
2826	 *    locally and then not have checked for the WriteDeny.)
2827	 */
2828	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2829	    stp = LIST_FIRST(&lfp->lf_deleg);
2830	    while (stp != LIST_END(&lfp->lf_deleg)) {
2831		nstp = LIST_NEXT(stp, ls_file);
2832		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2833			writedeleg = 0;
2834		else
2835			delegate = 0;
2836		if ((readonly && stp->ls_clp != clp &&
2837		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2838		    (!readonly && (stp->ls_clp != clp ||
2839		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2840		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2841			delegate = 2;
2842		    } else {
2843			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2844			if (ret) {
2845			    /*
2846			     * nfsrv_delegconflict() unlocks state
2847			     * when it returns non-zero.
2848			     */
2849			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2850			    free((caddr_t)new_open, M_NFSDSTATE);
2851			    free((caddr_t)new_deleg, M_NFSDSTATE);
2852			    if (ret == -1) {
2853				openstp = NULL;
2854				goto tryagain;
2855			    }
2856			    error = ret;
2857			    goto out;
2858			}
2859		    }
2860		}
2861		stp = nstp;
2862	    }
2863	}
2864
2865	/*
2866	 * We only get here if there was no open that conflicted.
2867	 * If an open for the owner exists, or in the access/deny bits.
2868	 * Otherwise it is a new open. If the open_owner hasn't been
2869	 * confirmed, replace the open with the new one needing confirmation,
2870	 * otherwise add the open.
2871	 */
2872	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2873	    /*
2874	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2875	     * a match. If found, just move the old delegation to the current
2876	     * delegation list and issue open. If not found, return
2877	     * NFSERR_EXPIRED.
2878	     */
2879	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2880		if (stp->ls_lfp == lfp) {
2881		    /* Found it */
2882		    if (stp->ls_clp != clp)
2883			panic("olddeleg clp");
2884		    LIST_REMOVE(stp, ls_list);
2885		    LIST_REMOVE(stp, ls_hash);
2886		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2887		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2888		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2889			clp->lc_clientid.lval[0];
2890		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2891			clp->lc_clientid.lval[1];
2892		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2893			nfsrv_nextstateindex(clp);
2894		    stp->ls_compref = nd->nd_compref;
2895		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2896		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2897			stp->ls_stateid), stp, ls_hash);
2898		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2899			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2900		    else
2901			*rflagsp |= NFSV4OPEN_READDELEGATE;
2902		    clp->lc_delegtime = NFSD_MONOSEC +
2903			nfsrv_lease + NFSRV_LEASEDELTA;
2904
2905		    /*
2906		     * Now, do the associated open.
2907		     */
2908		    new_open->ls_stateid.seqid = 1;
2909		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2910		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2911		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2912		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2913			NFSLCK_OPEN;
2914		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2915			new_open->ls_flags |= (NFSLCK_READACCESS |
2916			    NFSLCK_WRITEACCESS);
2917		    else
2918			new_open->ls_flags |= NFSLCK_READACCESS;
2919		    new_open->ls_uid = new_stp->ls_uid;
2920		    new_open->ls_lfp = lfp;
2921		    new_open->ls_clp = clp;
2922		    LIST_INIT(&new_open->ls_open);
2923		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2924		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2925			new_open, ls_hash);
2926		    /*
2927		     * and handle the open owner
2928		     */
2929		    if (ownerstp) {
2930			new_open->ls_openowner = ownerstp;
2931			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2932		    } else {
2933			new_open->ls_openowner = new_stp;
2934			new_stp->ls_flags = 0;
2935			nfsrvd_refcache(new_stp->ls_op);
2936			new_stp->ls_noopens = 0;
2937			LIST_INIT(&new_stp->ls_open);
2938			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2939			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2940			*new_stpp = NULL;
2941			newnfsstats.srvopenowners++;
2942			nfsrv_openpluslock++;
2943		    }
2944		    openstp = new_open;
2945		    new_open = NULL;
2946		    newnfsstats.srvopens++;
2947		    nfsrv_openpluslock++;
2948		    break;
2949		}
2950	    }
2951	    if (stp == LIST_END(&clp->lc_olddeleg))
2952		error = NFSERR_EXPIRED;
2953	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2954	    /*
2955	     * Scan to see that no delegation for this client and file
2956	     * doesn't already exist.
2957	     * There also shouldn't yet be an Open for this file and
2958	     * openowner.
2959	     */
2960	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2961		if (stp->ls_clp == clp)
2962		    break;
2963	    }
2964	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2965		/*
2966		 * This is the Claim_Previous case with a delegation
2967		 * type != Delegate_None.
2968		 */
2969		/*
2970		 * First, add the delegation. (Although we must issue the
2971		 * delegation, we can also ask for an immediate return.)
2972		 */
2973		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2974		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2975		    clp->lc_clientid.lval[0];
2976		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2977		    clp->lc_clientid.lval[1];
2978		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2979		    nfsrv_nextstateindex(clp);
2980		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2981		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2982			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2983		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2984		    nfsrv_writedelegcnt++;
2985		} else {
2986		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2987			NFSLCK_READACCESS);
2988		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2989		}
2990		new_deleg->ls_uid = new_stp->ls_uid;
2991		new_deleg->ls_lfp = lfp;
2992		new_deleg->ls_clp = clp;
2993		new_deleg->ls_filerev = filerev;
2994		new_deleg->ls_compref = nd->nd_compref;
2995		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2996		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2997		    new_deleg->ls_stateid), new_deleg, ls_hash);
2998		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2999		new_deleg = NULL;
3000		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3001		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3002		     LCL_CALLBACKSON ||
3003		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3004		    !NFSVNO_DELEGOK(vp))
3005		    *rflagsp |= NFSV4OPEN_RECALL;
3006		newnfsstats.srvdelegates++;
3007		nfsrv_openpluslock++;
3008		nfsrv_delegatecnt++;
3009
3010		/*
3011		 * Now, do the associated open.
3012		 */
3013		new_open->ls_stateid.seqid = 1;
3014		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3015		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3016		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3017		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3018		    NFSLCK_OPEN;
3019		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3020			new_open->ls_flags |= (NFSLCK_READACCESS |
3021			    NFSLCK_WRITEACCESS);
3022		else
3023			new_open->ls_flags |= NFSLCK_READACCESS;
3024		new_open->ls_uid = new_stp->ls_uid;
3025		new_open->ls_lfp = lfp;
3026		new_open->ls_clp = clp;
3027		LIST_INIT(&new_open->ls_open);
3028		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3029		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3030		   new_open, ls_hash);
3031		/*
3032		 * and handle the open owner
3033		 */
3034		if (ownerstp) {
3035		    new_open->ls_openowner = ownerstp;
3036		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3037		} else {
3038		    new_open->ls_openowner = new_stp;
3039		    new_stp->ls_flags = 0;
3040		    nfsrvd_refcache(new_stp->ls_op);
3041		    new_stp->ls_noopens = 0;
3042		    LIST_INIT(&new_stp->ls_open);
3043		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3044		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3045		    *new_stpp = NULL;
3046		    newnfsstats.srvopenowners++;
3047		    nfsrv_openpluslock++;
3048		}
3049		openstp = new_open;
3050		new_open = NULL;
3051		newnfsstats.srvopens++;
3052		nfsrv_openpluslock++;
3053	    } else {
3054		error = NFSERR_RECLAIMCONFLICT;
3055	    }
3056	} else if (ownerstp) {
3057		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3058		    /* Replace the open */
3059		    if (ownerstp->ls_op)
3060			nfsrvd_derefcache(ownerstp->ls_op);
3061		    ownerstp->ls_op = new_stp->ls_op;
3062		    nfsrvd_refcache(ownerstp->ls_op);
3063		    ownerstp->ls_seq = new_stp->ls_seq;
3064		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3065		    stp = LIST_FIRST(&ownerstp->ls_open);
3066		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3067			NFSLCK_OPEN;
3068		    stp->ls_stateid.seqid = 1;
3069		    stp->ls_uid = new_stp->ls_uid;
3070		    if (lfp != stp->ls_lfp) {
3071			LIST_REMOVE(stp, ls_file);
3072			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3073			stp->ls_lfp = lfp;
3074		    }
3075		    openstp = stp;
3076		} else if (openstp) {
3077		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3078		    openstp->ls_stateid.seqid++;
3079		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3080			openstp->ls_stateid.seqid == 0)
3081			openstp->ls_stateid.seqid = 1;
3082
3083		    /*
3084		     * This is where we can choose to issue a delegation.
3085		     */
3086		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3087			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3088		    else if (nfsrv_issuedelegs == 0)
3089			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3090		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3091			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3092		    else if (delegate == 0 || writedeleg == 0 ||
3093			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3094			nfsrv_writedelegifpos == 0) ||
3095			!NFSVNO_DELEGOK(vp) ||
3096			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3097			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3098			 LCL_CALLBACKSON)
3099			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3100		    else {
3101			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3102			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3103			    = clp->lc_clientid.lval[0];
3104			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3105			    = clp->lc_clientid.lval[1];
3106			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3107			    = nfsrv_nextstateindex(clp);
3108			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3109			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3110			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3111			new_deleg->ls_uid = new_stp->ls_uid;
3112			new_deleg->ls_lfp = lfp;
3113			new_deleg->ls_clp = clp;
3114			new_deleg->ls_filerev = filerev;
3115			new_deleg->ls_compref = nd->nd_compref;
3116			nfsrv_writedelegcnt++;
3117			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3118			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3119			    new_deleg->ls_stateid), new_deleg, ls_hash);
3120			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3121			new_deleg = NULL;
3122			newnfsstats.srvdelegates++;
3123			nfsrv_openpluslock++;
3124			nfsrv_delegatecnt++;
3125		    }
3126		} else {
3127		    new_open->ls_stateid.seqid = 1;
3128		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3129		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3130		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3131		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3132			NFSLCK_OPEN;
3133		    new_open->ls_uid = new_stp->ls_uid;
3134		    new_open->ls_openowner = ownerstp;
3135		    new_open->ls_lfp = lfp;
3136		    new_open->ls_clp = clp;
3137		    LIST_INIT(&new_open->ls_open);
3138		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3139		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3140		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3141			new_open, ls_hash);
3142		    openstp = new_open;
3143		    new_open = NULL;
3144		    newnfsstats.srvopens++;
3145		    nfsrv_openpluslock++;
3146
3147		    /*
3148		     * This is where we can choose to issue a delegation.
3149		     */
3150		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3151			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3152		    else if (nfsrv_issuedelegs == 0)
3153			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3154		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3155			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3156		    else if (delegate == 0 || (writedeleg == 0 &&
3157			readonly == 0) || !NFSVNO_DELEGOK(vp) ||
3158			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3159			 LCL_CALLBACKSON)
3160			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3161		    else {
3162			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3163			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3164			    = clp->lc_clientid.lval[0];
3165			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3166			    = clp->lc_clientid.lval[1];
3167			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3168			    = nfsrv_nextstateindex(clp);
3169			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3170			    (nfsrv_writedelegifpos || !readonly) &&
3171			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3172			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3173				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3174			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3175			    nfsrv_writedelegcnt++;
3176			} else {
3177			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3178				NFSLCK_READACCESS);
3179			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3180			}
3181			new_deleg->ls_uid = new_stp->ls_uid;
3182			new_deleg->ls_lfp = lfp;
3183			new_deleg->ls_clp = clp;
3184			new_deleg->ls_filerev = filerev;
3185			new_deleg->ls_compref = nd->nd_compref;
3186			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3187			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3188			    new_deleg->ls_stateid), new_deleg, ls_hash);
3189			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3190			new_deleg = NULL;
3191			newnfsstats.srvdelegates++;
3192			nfsrv_openpluslock++;
3193			nfsrv_delegatecnt++;
3194		    }
3195		}
3196	} else {
3197		/*
3198		 * New owner case. Start the open_owner sequence with a
3199		 * Needs confirmation (unless a reclaim) and hang the
3200		 * new open off it.
3201		 */
3202		new_open->ls_stateid.seqid = 1;
3203		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3204		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3205		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3206		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3207		    NFSLCK_OPEN;
3208		new_open->ls_uid = new_stp->ls_uid;
3209		LIST_INIT(&new_open->ls_open);
3210		new_open->ls_openowner = new_stp;
3211		new_open->ls_lfp = lfp;
3212		new_open->ls_clp = clp;
3213		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3214		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3215			new_stp->ls_flags = 0;
3216		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3217			/* NFSv4.1 never needs confirmation. */
3218			new_stp->ls_flags = 0;
3219
3220			/*
3221			 * This is where we can choose to issue a delegation.
3222			 */
3223			if (delegate && nfsrv_issuedelegs &&
3224			    (writedeleg || readonly) &&
3225			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3226			     LCL_CALLBACKSON &&
3227			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3228			    NFSVNO_DELEGOK(vp) &&
3229			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3230			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3231				new_deleg->ls_stateid.seqid =
3232				    delegstateidp->seqid = 1;
3233				new_deleg->ls_stateid.other[0] =
3234				    delegstateidp->other[0]
3235				    = clp->lc_clientid.lval[0];
3236				new_deleg->ls_stateid.other[1] =
3237				    delegstateidp->other[1]
3238				    = clp->lc_clientid.lval[1];
3239				new_deleg->ls_stateid.other[2] =
3240				    delegstateidp->other[2]
3241				    = nfsrv_nextstateindex(clp);
3242				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3243				    (nfsrv_writedelegifpos || !readonly) &&
3244				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3245				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3246				     0)) {
3247					new_deleg->ls_flags =
3248					    (NFSLCK_DELEGWRITE |
3249					     NFSLCK_READACCESS |
3250					     NFSLCK_WRITEACCESS);
3251					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3252					nfsrv_writedelegcnt++;
3253				} else {
3254					new_deleg->ls_flags =
3255					    (NFSLCK_DELEGREAD |
3256					     NFSLCK_READACCESS);
3257					*rflagsp |= NFSV4OPEN_READDELEGATE;
3258				}
3259				new_deleg->ls_uid = new_stp->ls_uid;
3260				new_deleg->ls_lfp = lfp;
3261				new_deleg->ls_clp = clp;
3262				new_deleg->ls_filerev = filerev;
3263				new_deleg->ls_compref = nd->nd_compref;
3264				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3265				    ls_file);
3266				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3267				    new_deleg->ls_stateid), new_deleg, ls_hash);
3268				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3269				    ls_list);
3270				new_deleg = NULL;
3271				newnfsstats.srvdelegates++;
3272				nfsrv_openpluslock++;
3273				nfsrv_delegatecnt++;
3274			}
3275			/*
3276			 * Since NFSv4.1 never does an OpenConfirm, the first
3277			 * open state will be acquired here.
3278			 */
3279			if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3280				clp->lc_flags |= LCL_STAMPEDSTABLE;
3281				len = clp->lc_idlen;
3282				NFSBCOPY(clp->lc_id, clidp, len);
3283				gotstate = 1;
3284			}
3285		} else {
3286			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3287			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3288		}
3289		nfsrvd_refcache(new_stp->ls_op);
3290		new_stp->ls_noopens = 0;
3291		LIST_INIT(&new_stp->ls_open);
3292		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3293		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3294		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3295		    new_open, ls_hash);
3296		openstp = new_open;
3297		new_open = NULL;
3298		*new_stpp = NULL;
3299		newnfsstats.srvopens++;
3300		nfsrv_openpluslock++;
3301		newnfsstats.srvopenowners++;
3302		nfsrv_openpluslock++;
3303	}
3304	if (!error) {
3305		stateidp->seqid = openstp->ls_stateid.seqid;
3306		stateidp->other[0] = openstp->ls_stateid.other[0];
3307		stateidp->other[1] = openstp->ls_stateid.other[1];
3308		stateidp->other[2] = openstp->ls_stateid.other[2];
3309	}
3310	NFSUNLOCKSTATE();
3311	if (haslock) {
3312		NFSLOCKV4ROOTMUTEX();
3313		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3314		NFSUNLOCKV4ROOTMUTEX();
3315	}
3316	if (new_open)
3317		FREE((caddr_t)new_open, M_NFSDSTATE);
3318	if (new_deleg)
3319		FREE((caddr_t)new_deleg, M_NFSDSTATE);
3320
3321	/*
3322	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3323	 * to the stable storage file.
3324	 */
3325	if (gotstate != 0) {
3326		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3327		nfsrv_backupstable();
3328	}
3329
3330out:
3331	free(clidp, M_TEMP);
3332	NFSEXITCODE2(error, nd);
3333	return (error);
3334}
3335
3336/*
3337 * Open update. Does the confirm, downgrade and close.
3338 */
3339APPLESTATIC int
3340nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3341    nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
3342{
3343	struct nfsstate *stp, *ownerstp;
3344	struct nfsclient *clp;
3345	struct nfslockfile *lfp;
3346	u_int32_t bits;
3347	int error = 0, gotstate = 0, len = 0;
3348	u_char *clidp = NULL;
3349
3350	/*
3351	 * Check for restart conditions (client and server).
3352	 */
3353	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3354	    &new_stp->ls_stateid, 0);
3355	if (error)
3356		goto out;
3357
3358	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3359	NFSLOCKSTATE();
3360	/*
3361	 * Get the open structure via clientid and stateid.
3362	 */
3363	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3364	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3365	if (!error)
3366		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3367		    new_stp->ls_flags, &stp);
3368
3369	/*
3370	 * Sanity check the open.
3371	 */
3372	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3373		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3374		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3375		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3376		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3377		error = NFSERR_BADSTATEID;
3378
3379	if (!error)
3380		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3381		    stp->ls_openowner, new_stp->ls_op);
3382	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3383	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3384	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3385	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3386	      new_stp->ls_stateid.seqid != 0)))
3387		error = NFSERR_OLDSTATEID;
3388	if (!error && vnode_vtype(vp) != VREG) {
3389		if (vnode_vtype(vp) == VDIR)
3390			error = NFSERR_ISDIR;
3391		else
3392			error = NFSERR_INVAL;
3393	}
3394
3395	if (error) {
3396		/*
3397		 * If a client tries to confirm an Open with a bad
3398		 * seqid# and there are no byte range locks or other Opens
3399		 * on the openowner, just throw it away, so the next use of the
3400		 * openowner will start a fresh seq#.
3401		 */
3402		if (error == NFSERR_BADSEQID &&
3403		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3404		    nfsrv_nootherstate(stp))
3405			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3406		NFSUNLOCKSTATE();
3407		goto out;
3408	}
3409
3410	/*
3411	 * Set the return stateid.
3412	 */
3413	stateidp->seqid = stp->ls_stateid.seqid + 1;
3414	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3415		stateidp->seqid = 1;
3416	stateidp->other[0] = stp->ls_stateid.other[0];
3417	stateidp->other[1] = stp->ls_stateid.other[1];
3418	stateidp->other[2] = stp->ls_stateid.other[2];
3419	/*
3420	 * Now, handle the three cases.
3421	 */
3422	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3423		/*
3424		 * If the open doesn't need confirmation, it seems to me that
3425		 * there is a client error, but I'll just log it and keep going?
3426		 */
3427		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3428			printf("Nfsv4d: stray open confirm\n");
3429		stp->ls_openowner->ls_flags = 0;
3430		stp->ls_stateid.seqid++;
3431		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3432		    stp->ls_stateid.seqid == 0)
3433			stp->ls_stateid.seqid = 1;
3434		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3435			clp->lc_flags |= LCL_STAMPEDSTABLE;
3436			len = clp->lc_idlen;
3437			NFSBCOPY(clp->lc_id, clidp, len);
3438			gotstate = 1;
3439		}
3440		NFSUNLOCKSTATE();
3441	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3442		ownerstp = stp->ls_openowner;
3443		lfp = stp->ls_lfp;
3444		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3445			/* Get the lf lock */
3446			nfsrv_locklf(lfp);
3447			NFSUNLOCKSTATE();
3448			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3449			NFSVOPUNLOCK(vp, 0);
3450			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3451				NFSLOCKSTATE();
3452				nfsrv_unlocklf(lfp);
3453				NFSUNLOCKSTATE();
3454			}
3455			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3456		} else {
3457			(void) nfsrv_freeopen(stp, NULL, 0, p);
3458			NFSUNLOCKSTATE();
3459		}
3460	} else {
3461		/*
3462		 * Update the share bits, making sure that the new set are a
3463		 * subset of the old ones.
3464		 */
3465		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3466		if (~(stp->ls_flags) & bits) {
3467			NFSUNLOCKSTATE();
3468			error = NFSERR_INVAL;
3469			goto out;
3470		}
3471		stp->ls_flags = (bits | NFSLCK_OPEN);
3472		stp->ls_stateid.seqid++;
3473		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3474		    stp->ls_stateid.seqid == 0)
3475			stp->ls_stateid.seqid = 1;
3476		NFSUNLOCKSTATE();
3477	}
3478
3479	/*
3480	 * If the client just confirmed its first open, write a timestamp
3481	 * to the stable storage file.
3482	 */
3483	if (gotstate != 0) {
3484		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3485		nfsrv_backupstable();
3486	}
3487
3488out:
3489	free(clidp, M_TEMP);
3490	NFSEXITCODE2(error, nd);
3491	return (error);
3492}
3493
3494/*
3495 * Delegation update. Does the purge and return.
3496 */
3497APPLESTATIC int
3498nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3499    nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3500    NFSPROC_T *p)
3501{
3502	struct nfsstate *stp;
3503	struct nfsclient *clp;
3504	int error = 0;
3505	fhandle_t fh;
3506
3507	/*
3508	 * Do a sanity check against the file handle for DelegReturn.
3509	 */
3510	if (vp) {
3511		error = nfsvno_getfh(vp, &fh, p);
3512		if (error)
3513			goto out;
3514	}
3515	/*
3516	 * Check for restart conditions (client and server).
3517	 */
3518	if (op == NFSV4OP_DELEGRETURN)
3519		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3520			stateidp, 0);
3521	else
3522		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3523			stateidp, 0);
3524
3525	NFSLOCKSTATE();
3526	/*
3527	 * Get the open structure via clientid and stateid.
3528	 */
3529	if (!error)
3530	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3531		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3532	if (error) {
3533		if (error == NFSERR_CBPATHDOWN)
3534			error = 0;
3535		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3536			error = NFSERR_STALESTATEID;
3537	}
3538	if (!error && op == NFSV4OP_DELEGRETURN) {
3539	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3540	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3541		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3542		error = NFSERR_OLDSTATEID;
3543	}
3544	/*
3545	 * NFSERR_EXPIRED means that the state has gone away,
3546	 * so Delegations have been purged. Just return ok.
3547	 */
3548	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3549		NFSUNLOCKSTATE();
3550		error = 0;
3551		goto out;
3552	}
3553	if (error) {
3554		NFSUNLOCKSTATE();
3555		goto out;
3556	}
3557
3558	if (op == NFSV4OP_DELEGRETURN) {
3559		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3560		    sizeof (fhandle_t))) {
3561			NFSUNLOCKSTATE();
3562			error = NFSERR_BADSTATEID;
3563			goto out;
3564		}
3565		nfsrv_freedeleg(stp);
3566	} else {
3567		nfsrv_freedeleglist(&clp->lc_olddeleg);
3568	}
3569	NFSUNLOCKSTATE();
3570	error = 0;
3571
3572out:
3573	NFSEXITCODE(error);
3574	return (error);
3575}
3576
3577/*
3578 * Release lock owner.
3579 */
3580APPLESTATIC int
3581nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3582    NFSPROC_T *p)
3583{
3584	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3585	struct nfsclient *clp;
3586	int error = 0;
3587
3588	/*
3589	 * Check for restart conditions (client and server).
3590	 */
3591	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3592	    &new_stp->ls_stateid, 0);
3593	if (error)
3594		goto out;
3595
3596	NFSLOCKSTATE();
3597	/*
3598	 * Get the lock owner by name.
3599	 */
3600	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3601	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3602	if (error) {
3603		NFSUNLOCKSTATE();
3604		goto out;
3605	}
3606	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3607	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3608		stp = LIST_FIRST(&openstp->ls_open);
3609		while (stp != LIST_END(&openstp->ls_open)) {
3610		    nstp = LIST_NEXT(stp, ls_list);
3611		    /*
3612		     * If the owner matches, check for locks and
3613		     * then free or return an error.
3614		     */
3615		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3616			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3617			 stp->ls_ownerlen)){
3618			if (LIST_EMPTY(&stp->ls_lock)) {
3619			    nfsrv_freelockowner(stp, NULL, 0, p);
3620			} else {
3621			    NFSUNLOCKSTATE();
3622			    error = NFSERR_LOCKSHELD;
3623			    goto out;
3624			}
3625		    }
3626		    stp = nstp;
3627		}
3628	    }
3629	}
3630	NFSUNLOCKSTATE();
3631
3632out:
3633	NFSEXITCODE(error);
3634	return (error);
3635}
3636
3637/*
3638 * Get the file handle for a lock structure.
3639 */
3640static int
3641nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3642    fhandle_t *nfhp, NFSPROC_T *p)
3643{
3644	fhandle_t *fhp = NULL;
3645	int error;
3646
3647	/*
3648	 * For lock, use the new nfslock structure, otherwise just
3649	 * a fhandle_t on the stack.
3650	 */
3651	if (flags & NFSLCK_OPEN) {
3652		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3653		fhp = &new_lfp->lf_fh;
3654	} else if (nfhp) {
3655		fhp = nfhp;
3656	} else {
3657		panic("nfsrv_getlockfh");
3658	}
3659	error = nfsvno_getfh(vp, fhp, p);
3660	NFSEXITCODE(error);
3661	return (error);
3662}
3663
3664/*
3665 * Get an nfs lock structure. Allocate one, as required, and return a
3666 * pointer to it.
3667 * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3668 */
3669static int
3670nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3671    struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3672{
3673	struct nfslockfile *lfp;
3674	fhandle_t *fhp = NULL, *tfhp;
3675	struct nfslockhashhead *hp;
3676	struct nfslockfile *new_lfp = NULL;
3677
3678	/*
3679	 * For lock, use the new nfslock structure, otherwise just
3680	 * a fhandle_t on the stack.
3681	 */
3682	if (flags & NFSLCK_OPEN) {
3683		new_lfp = *new_lfpp;
3684		fhp = &new_lfp->lf_fh;
3685	} else if (nfhp) {
3686		fhp = nfhp;
3687	} else {
3688		panic("nfsrv_getlockfile");
3689	}
3690
3691	hp = NFSLOCKHASH(fhp);
3692	LIST_FOREACH(lfp, hp, lf_hash) {
3693		tfhp = &lfp->lf_fh;
3694		if (NFSVNO_CMPFH(fhp, tfhp)) {
3695			if (lockit)
3696				nfsrv_locklf(lfp);
3697			*lfpp = lfp;
3698			return (0);
3699		}
3700	}
3701	if (!(flags & NFSLCK_OPEN))
3702		return (-1);
3703
3704	/*
3705	 * No match, so chain the new one into the list.
3706	 */
3707	LIST_INIT(&new_lfp->lf_open);
3708	LIST_INIT(&new_lfp->lf_lock);
3709	LIST_INIT(&new_lfp->lf_deleg);
3710	LIST_INIT(&new_lfp->lf_locallock);
3711	LIST_INIT(&new_lfp->lf_rollback);
3712	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3713	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3714	new_lfp->lf_usecount = 0;
3715	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3716	*lfpp = new_lfp;
3717	*new_lfpp = NULL;
3718	return (0);
3719}
3720
3721/*
3722 * This function adds a nfslock lock structure to the list for the associated
3723 * nfsstate and nfslockfile structures. It will be inserted after the
3724 * entry pointed at by insert_lop.
3725 */
3726static void
3727nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3728    struct nfsstate *stp, struct nfslockfile *lfp)
3729{
3730	struct nfslock *lop, *nlop;
3731
3732	new_lop->lo_stp = stp;
3733	new_lop->lo_lfp = lfp;
3734
3735	if (stp != NULL) {
3736		/* Insert in increasing lo_first order */
3737		lop = LIST_FIRST(&lfp->lf_lock);
3738		if (lop == LIST_END(&lfp->lf_lock) ||
3739		    new_lop->lo_first <= lop->lo_first) {
3740			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3741		} else {
3742			nlop = LIST_NEXT(lop, lo_lckfile);
3743			while (nlop != LIST_END(&lfp->lf_lock) &&
3744			       nlop->lo_first < new_lop->lo_first) {
3745				lop = nlop;
3746				nlop = LIST_NEXT(lop, lo_lckfile);
3747			}
3748			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3749		}
3750	} else {
3751		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3752	}
3753
3754	/*
3755	 * Insert after insert_lop, which is overloaded as stp or lfp for
3756	 * an empty list.
3757	 */
3758	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3759		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3760	else if ((struct nfsstate *)insert_lop == stp)
3761		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3762	else
3763		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3764	if (stp != NULL) {
3765		newnfsstats.srvlocks++;
3766		nfsrv_openpluslock++;
3767	}
3768}
3769
3770/*
3771 * This function updates the locking for a lock owner and given file. It
3772 * maintains a list of lock ranges ordered on increasing file offset that
3773 * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3774 * It always adds new_lop to the list and sometimes uses the one pointed
3775 * at by other_lopp.
3776 */
3777static void
3778nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3779    struct nfslock **other_lopp, struct nfslockfile *lfp)
3780{
3781	struct nfslock *new_lop = *new_lopp;
3782	struct nfslock *lop, *tlop, *ilop;
3783	struct nfslock *other_lop = *other_lopp;
3784	int unlock = 0, myfile = 0;
3785	u_int64_t tmp;
3786
3787	/*
3788	 * Work down the list until the lock is merged.
3789	 */
3790	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3791		unlock = 1;
3792	if (stp != NULL) {
3793		ilop = (struct nfslock *)stp;
3794		lop = LIST_FIRST(&stp->ls_lock);
3795	} else {
3796		ilop = (struct nfslock *)lfp;
3797		lop = LIST_FIRST(&lfp->lf_locallock);
3798	}
3799	while (lop != NULL) {
3800	    /*
3801	     * Only check locks for this file that aren't before the start of
3802	     * new lock's range.
3803	     */
3804	    if (lop->lo_lfp == lfp) {
3805	      myfile = 1;
3806	      if (lop->lo_end >= new_lop->lo_first) {
3807		if (new_lop->lo_end < lop->lo_first) {
3808			/*
3809			 * If the new lock ends before the start of the
3810			 * current lock's range, no merge, just insert
3811			 * the new lock.
3812			 */
3813			break;
3814		}
3815		if (new_lop->lo_flags == lop->lo_flags ||
3816		    (new_lop->lo_first <= lop->lo_first &&
3817		     new_lop->lo_end >= lop->lo_end)) {
3818			/*
3819			 * This lock can be absorbed by the new lock/unlock.
3820			 * This happens when it covers the entire range
3821			 * of the old lock or is contiguous
3822			 * with the old lock and is of the same type or an
3823			 * unlock.
3824			 */
3825			if (lop->lo_first < new_lop->lo_first)
3826				new_lop->lo_first = lop->lo_first;
3827			if (lop->lo_end > new_lop->lo_end)
3828				new_lop->lo_end = lop->lo_end;
3829			tlop = lop;
3830			lop = LIST_NEXT(lop, lo_lckowner);
3831			nfsrv_freenfslock(tlop);
3832			continue;
3833		}
3834
3835		/*
3836		 * All these cases are for contiguous locks that are not the
3837		 * same type, so they can't be merged.
3838		 */
3839		if (new_lop->lo_first <= lop->lo_first) {
3840			/*
3841			 * This case is where the new lock overlaps with the
3842			 * first part of the old lock. Move the start of the
3843			 * old lock to just past the end of the new lock. The
3844			 * new lock will be inserted in front of the old, since
3845			 * ilop hasn't been updated. (We are done now.)
3846			 */
3847			lop->lo_first = new_lop->lo_end;
3848			break;
3849		}
3850		if (new_lop->lo_end >= lop->lo_end) {
3851			/*
3852			 * This case is where the new lock overlaps with the
3853			 * end of the old lock's range. Move the old lock's
3854			 * end to just before the new lock's first and insert
3855			 * the new lock after the old lock.
3856			 * Might not be done yet, since the new lock could
3857			 * overlap further locks with higher ranges.
3858			 */
3859			lop->lo_end = new_lop->lo_first;
3860			ilop = lop;
3861			lop = LIST_NEXT(lop, lo_lckowner);
3862			continue;
3863		}
3864		/*
3865		 * The final case is where the new lock's range is in the
3866		 * middle of the current lock's and splits the current lock
3867		 * up. Use *other_lopp to handle the second part of the
3868		 * split old lock range. (We are done now.)
3869		 * For unlock, we use new_lop as other_lop and tmp, since
3870		 * other_lop and new_lop are the same for this case.
3871		 * We noted the unlock case above, so we don't need
3872		 * new_lop->lo_flags any longer.
3873		 */
3874		tmp = new_lop->lo_first;
3875		if (other_lop == NULL) {
3876			if (!unlock)
3877				panic("nfsd srv update unlock");
3878			other_lop = new_lop;
3879			*new_lopp = NULL;
3880		}
3881		other_lop->lo_first = new_lop->lo_end;
3882		other_lop->lo_end = lop->lo_end;
3883		other_lop->lo_flags = lop->lo_flags;
3884		other_lop->lo_stp = stp;
3885		other_lop->lo_lfp = lfp;
3886		lop->lo_end = tmp;
3887		nfsrv_insertlock(other_lop, lop, stp, lfp);
3888		*other_lopp = NULL;
3889		ilop = lop;
3890		break;
3891	      }
3892	    }
3893	    ilop = lop;
3894	    lop = LIST_NEXT(lop, lo_lckowner);
3895	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3896		break;
3897	}
3898
3899	/*
3900	 * Insert the new lock in the list at the appropriate place.
3901	 */
3902	if (!unlock) {
3903		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3904		*new_lopp = NULL;
3905	}
3906}
3907
3908/*
3909 * This function handles sequencing of locks, etc.
3910 * It returns an error that indicates what the caller should do.
3911 */
3912static int
3913nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3914    struct nfsstate *stp, struct nfsrvcache *op)
3915{
3916	int error = 0;
3917
3918	if ((nd->nd_flag & ND_NFSV41) != 0)
3919		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
3920		goto out;
3921	if (op != nd->nd_rp)
3922		panic("nfsrvstate checkseqid");
3923	if (!(op->rc_flag & RC_INPROG))
3924		panic("nfsrvstate not inprog");
3925	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3926		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3927		panic("nfsrvstate op refcnt");
3928	}
3929	if ((stp->ls_seq + 1) == seqid) {
3930		if (stp->ls_op)
3931			nfsrvd_derefcache(stp->ls_op);
3932		stp->ls_op = op;
3933		nfsrvd_refcache(op);
3934		stp->ls_seq = seqid;
3935		goto out;
3936	} else if (stp->ls_seq == seqid && stp->ls_op &&
3937		op->rc_xid == stp->ls_op->rc_xid &&
3938		op->rc_refcnt == 0 &&
3939		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3940		op->rc_cksum == stp->ls_op->rc_cksum) {
3941		if (stp->ls_op->rc_flag & RC_INPROG) {
3942			error = NFSERR_DONTREPLY;
3943			goto out;
3944		}
3945		nd->nd_rp = stp->ls_op;
3946		nd->nd_rp->rc_flag |= RC_INPROG;
3947		nfsrvd_delcache(op);
3948		error = NFSERR_REPLYFROMCACHE;
3949		goto out;
3950	}
3951	error = NFSERR_BADSEQID;
3952
3953out:
3954	NFSEXITCODE2(error, nd);
3955	return (error);
3956}
3957
3958/*
3959 * Get the client ip address for callbacks. If the strings can't be parsed,
3960 * just set lc_program to 0 to indicate no callbacks are possible.
3961 * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3962 *  the address to the client's transport address. This won't be used
3963 *  for callbacks, but can be printed out by newnfsstats for info.)
3964 * Return error if the xdr can't be parsed, 0 otherwise.
3965 */
3966APPLESTATIC int
3967nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3968{
3969	u_int32_t *tl;
3970	u_char *cp, *cp2;
3971	int i, j, maxalen = 0, minalen = 0;
3972	sa_family_t af;
3973#ifdef INET
3974	struct sockaddr_in *rin, *sin;
3975#endif
3976#ifdef INET6
3977	struct sockaddr_in6 *rin6, *sin6;
3978#endif
3979	u_char *addr;
3980	int error = 0, cantparse = 0;
3981	union {
3982		in_addr_t ival;
3983		u_char cval[4];
3984	} ip;
3985	union {
3986		in_port_t sval;
3987		u_char cval[2];
3988	} port;
3989
3990	/* 8 is the maximum length of the port# string. */
3991	addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK);
3992	clp->lc_req.nr_client = NULL;
3993	clp->lc_req.nr_lock = 0;
3994	af = AF_UNSPEC;
3995	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3996	i = fxdr_unsigned(int, *tl);
3997	if (i >= 3 && i <= 4) {
3998		error = nfsrv_mtostr(nd, addr, i);
3999		if (error)
4000			goto nfsmout;
4001#ifdef INET
4002		if (!strcmp(addr, "tcp")) {
4003			clp->lc_flags |= LCL_TCPCALLBACK;
4004			clp->lc_req.nr_sotype = SOCK_STREAM;
4005			clp->lc_req.nr_soproto = IPPROTO_TCP;
4006			af = AF_INET;
4007		} else if (!strcmp(addr, "udp")) {
4008			clp->lc_req.nr_sotype = SOCK_DGRAM;
4009			clp->lc_req.nr_soproto = IPPROTO_UDP;
4010			af = AF_INET;
4011		}
4012#endif
4013#ifdef INET6
4014		if (af == AF_UNSPEC) {
4015			if (!strcmp(addr, "tcp6")) {
4016				clp->lc_flags |= LCL_TCPCALLBACK;
4017				clp->lc_req.nr_sotype = SOCK_STREAM;
4018				clp->lc_req.nr_soproto = IPPROTO_TCP;
4019				af = AF_INET6;
4020			} else if (!strcmp(addr, "udp6")) {
4021				clp->lc_req.nr_sotype = SOCK_DGRAM;
4022				clp->lc_req.nr_soproto = IPPROTO_UDP;
4023				af = AF_INET6;
4024			}
4025		}
4026#endif
4027		if (af == AF_UNSPEC) {
4028			cantparse = 1;
4029		}
4030	} else {
4031		cantparse = 1;
4032		if (i > 0) {
4033			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4034			if (error)
4035				goto nfsmout;
4036		}
4037	}
4038	/*
4039	 * The caller has allocated clp->lc_req.nr_nam to be large enough
4040	 * for either AF_INET or AF_INET6 and zeroed out the contents.
4041	 * maxalen is set to the maximum length of the host IP address string
4042	 * plus 8 for the maximum length of the port#.
4043	 * minalen is set to the minimum length of the host IP address string
4044	 * plus 4 for the minimum length of the port#.
4045	 * These lengths do not include NULL termination,
4046	 * so INET[6]_ADDRSTRLEN - 1 is used in the calculations.
4047	 */
4048	switch (af) {
4049#ifdef INET
4050	case AF_INET:
4051		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4052		rin->sin_family = AF_INET;
4053		rin->sin_len = sizeof(struct sockaddr_in);
4054		maxalen = INET_ADDRSTRLEN - 1 + 8;
4055		minalen = 7 + 4;
4056		break;
4057#endif
4058#ifdef INET6
4059	case AF_INET6:
4060		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4061		rin6->sin6_family = AF_INET6;
4062		rin6->sin6_len = sizeof(struct sockaddr_in6);
4063		maxalen = INET6_ADDRSTRLEN - 1 + 8;
4064		minalen = 3 + 4;
4065		break;
4066#endif
4067	}
4068	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4069	i = fxdr_unsigned(int, *tl);
4070	if (i < 0) {
4071		error = NFSERR_BADXDR;
4072		goto nfsmout;
4073	} else if (i == 0) {
4074		cantparse = 1;
4075	} else if (!cantparse && i <= maxalen && i >= minalen) {
4076		error = nfsrv_mtostr(nd, addr, i);
4077		if (error)
4078			goto nfsmout;
4079
4080		/*
4081		 * Parse out the address fields. We expect 6 decimal numbers
4082		 * separated by '.'s for AF_INET and two decimal numbers
4083		 * preceeded by '.'s for AF_INET6.
4084		 */
4085		cp = NULL;
4086		switch (af) {
4087#ifdef INET6
4088		/*
4089		 * For AF_INET6, first parse the host address.
4090		 */
4091		case AF_INET6:
4092			cp = strchr(addr, '.');
4093			if (cp != NULL) {
4094				*cp++ = '\0';
4095				if (inet_pton(af, addr, &rin6->sin6_addr) == 1)
4096					i = 4;
4097				else {
4098					cp = NULL;
4099					cantparse = 1;
4100				}
4101			}
4102			break;
4103#endif
4104#ifdef INET
4105		case AF_INET:
4106			cp = addr;
4107			i = 0;
4108			break;
4109#endif
4110		}
4111		while (cp != NULL && *cp && i < 6) {
4112			cp2 = cp;
4113			while (*cp2 && *cp2 != '.')
4114				cp2++;
4115			if (*cp2)
4116				*cp2++ = '\0';
4117			else if (i != 5) {
4118				cantparse = 1;
4119				break;
4120			}
4121			j = nfsrv_getipnumber(cp);
4122			if (j >= 0) {
4123				if (i < 4)
4124					ip.cval[3 - i] = j;
4125				else
4126					port.cval[5 - i] = j;
4127			} else {
4128				cantparse = 1;
4129				break;
4130			}
4131			cp = cp2;
4132			i++;
4133		}
4134		if (!cantparse) {
4135			/*
4136			 * The host address INADDR_ANY is (mis)used to indicate
4137			 * "there is no valid callback address".
4138			 */
4139			switch (af) {
4140#ifdef INET6
4141			case AF_INET6:
4142				if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr,
4143				    &in6addr_any))
4144					rin6->sin6_port = htons(port.sval);
4145				else
4146					cantparse = 1;
4147				break;
4148#endif
4149#ifdef INET
4150			case AF_INET:
4151				if (ip.ival != INADDR_ANY) {
4152					rin->sin_addr.s_addr = htonl(ip.ival);
4153					rin->sin_port = htons(port.sval);
4154				} else {
4155					cantparse = 1;
4156				}
4157				break;
4158#endif
4159			}
4160		}
4161	} else {
4162		cantparse = 1;
4163		if (i > 0) {
4164			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4165			if (error)
4166				goto nfsmout;
4167		}
4168	}
4169	if (cantparse) {
4170		switch (nd->nd_nam->sa_family) {
4171#ifdef INET
4172		case AF_INET:
4173			sin = (struct sockaddr_in *)nd->nd_nam;
4174			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4175			rin->sin_family = AF_INET;
4176			rin->sin_len = sizeof(struct sockaddr_in);
4177			rin->sin_addr.s_addr = sin->sin_addr.s_addr;
4178			rin->sin_port = 0x0;
4179			break;
4180#endif
4181#ifdef INET6
4182		case AF_INET6:
4183			sin6 = (struct sockaddr_in6 *)nd->nd_nam;
4184			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4185			rin6->sin6_family = AF_INET6;
4186			rin6->sin6_len = sizeof(struct sockaddr_in6);
4187			rin6->sin6_addr = sin6->sin6_addr;
4188			rin6->sin6_port = 0x0;
4189			break;
4190#endif
4191		}
4192		clp->lc_program = 0;
4193	}
4194nfsmout:
4195	free(addr, M_TEMP);
4196	NFSEXITCODE2(error, nd);
4197	return (error);
4198}
4199
4200/*
4201 * Turn a string of up to three decimal digits into a number. Return -1 upon
4202 * error.
4203 */
4204static int
4205nfsrv_getipnumber(u_char *cp)
4206{
4207	int i = 0, j = 0;
4208
4209	while (*cp) {
4210		if (j > 2 || *cp < '0' || *cp > '9')
4211			return (-1);
4212		i *= 10;
4213		i += (*cp - '0');
4214		cp++;
4215		j++;
4216	}
4217	if (i < 256)
4218		return (i);
4219	return (-1);
4220}
4221
4222/*
4223 * This function checks for restart conditions.
4224 */
4225static int
4226nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4227    nfsv4stateid_t *stateidp, int specialid)
4228{
4229	int ret = 0;
4230
4231	/*
4232	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4233	 * and DelegPurge have a clientid, the rest a stateid.
4234	 */
4235	if (flags &
4236	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4237		if (clientid.lval[0] != nfsrvboottime) {
4238			ret = NFSERR_STALECLIENTID;
4239			goto out;
4240		}
4241	} else if (stateidp->other[0] != nfsrvboottime &&
4242		specialid == 0) {
4243		ret = NFSERR_STALESTATEID;
4244		goto out;
4245	}
4246
4247	/*
4248	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4249	 * not use a lock/open owner seqid#, so the check can be done now.
4250	 * (The others will be checked, as required, later.)
4251	 */
4252	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4253		goto out;
4254
4255	NFSLOCKSTATE();
4256	ret = nfsrv_checkgrace(NULL, NULL, flags);
4257	NFSUNLOCKSTATE();
4258
4259out:
4260	NFSEXITCODE(ret);
4261	return (ret);
4262}
4263
4264/*
4265 * Check for grace.
4266 */
4267static int
4268nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4269    u_int32_t flags)
4270{
4271	int error = 0, notreclaimed;
4272	struct nfsrv_stable *sp;
4273
4274	if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE |
4275	     NFSNSF_GRACEOVER)) == 0) {
4276		/*
4277		 * First, check to see if all of the clients have done a
4278		 * ReclaimComplete.  If so, grace can end now.
4279		 */
4280		notreclaimed = 0;
4281		LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4282			if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4283				notreclaimed = 1;
4284				break;
4285			}
4286		}
4287		if (notreclaimed == 0)
4288			nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER |
4289			    NFSNSF_NEEDLOCK);
4290	}
4291
4292	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4293		if (flags & NFSLCK_RECLAIM) {
4294			error = NFSERR_NOGRACE;
4295			goto out;
4296		}
4297	} else {
4298		if (!(flags & NFSLCK_RECLAIM)) {
4299			error = NFSERR_GRACE;
4300			goto out;
4301		}
4302		if (nd != NULL && clp != NULL &&
4303		    (nd->nd_flag & ND_NFSV41) != 0 &&
4304		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4305			error = NFSERR_NOGRACE;
4306			goto out;
4307		}
4308
4309		/*
4310		 * If grace is almost over and we are still getting Reclaims,
4311		 * extend grace a bit.
4312		 */
4313		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4314		    nfsrv_stablefirst.nsf_eograce)
4315			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4316				NFSRV_LEASEDELTA;
4317	}
4318
4319out:
4320	NFSEXITCODE(error);
4321	return (error);
4322}
4323
4324/*
4325 * Do a server callback.
4326 */
4327static int
4328nfsrv_docallback(struct nfsclient *clp, int procnum,
4329    nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
4330    struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
4331{
4332	mbuf_t m;
4333	u_int32_t *tl;
4334	struct nfsrv_descript nfsd, *nd = &nfsd;
4335	struct ucred *cred;
4336	int error = 0;
4337	u_int32_t callback;
4338	struct nfsdsession *sep = NULL;
4339
4340	cred = newnfs_getcred();
4341	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4342	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4343		NFSUNLOCKSTATE();
4344		panic("docallb");
4345	}
4346	clp->lc_cbref++;
4347
4348	/*
4349	 * Fill the callback program# and version into the request
4350	 * structure for newnfs_connect() to use.
4351	 */
4352	clp->lc_req.nr_prog = clp->lc_program;
4353#ifdef notnow
4354	if ((clp->lc_flags & LCL_NFSV41) != 0)
4355		clp->lc_req.nr_vers = NFSV41_CBVERS;
4356	else
4357#endif
4358		clp->lc_req.nr_vers = NFSV4_CBVERS;
4359
4360	/*
4361	 * First, fill in some of the fields of nd and cr.
4362	 */
4363	nd->nd_flag = ND_NFSV4;
4364	if (clp->lc_flags & LCL_GSS)
4365		nd->nd_flag |= ND_KERBV;
4366	if ((clp->lc_flags & LCL_NFSV41) != 0)
4367		nd->nd_flag |= ND_NFSV41;
4368	nd->nd_repstat = 0;
4369	cred->cr_uid = clp->lc_uid;
4370	cred->cr_gid = clp->lc_gid;
4371	callback = clp->lc_callback;
4372	NFSUNLOCKSTATE();
4373	cred->cr_ngroups = 1;
4374
4375	/*
4376	 * Get the first mbuf for the request.
4377	 */
4378	MGET(m, M_WAITOK, MT_DATA);
4379	mbuf_setlen(m, 0);
4380	nd->nd_mreq = nd->nd_mb = m;
4381	nd->nd_bpos = NFSMTOD(m, caddr_t);
4382
4383	/*
4384	 * and build the callback request.
4385	 */
4386	if (procnum == NFSV4OP_CBGETATTR) {
4387		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4388		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4389		    "CB Getattr", &sep);
4390		if (error != 0) {
4391			mbuf_freem(nd->nd_mreq);
4392			goto errout;
4393		}
4394		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4395		(void)nfsrv_putattrbit(nd, attrbitp);
4396	} else if (procnum == NFSV4OP_CBRECALL) {
4397		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4398		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4399		    "CB Recall", &sep);
4400		if (error != 0) {
4401			mbuf_freem(nd->nd_mreq);
4402			goto errout;
4403		}
4404		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4405		*tl++ = txdr_unsigned(stateidp->seqid);
4406		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4407		    NFSX_STATEIDOTHER);
4408		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4409		if (trunc)
4410			*tl = newnfs_true;
4411		else
4412			*tl = newnfs_false;
4413		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4414	} else if (procnum == NFSV4PROC_CBNULL) {
4415		nd->nd_procnum = NFSV4PROC_CBNULL;
4416		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4417			error = nfsv4_getcbsession(clp, &sep);
4418			if (error != 0) {
4419				mbuf_freem(nd->nd_mreq);
4420				goto errout;
4421			}
4422		}
4423	} else {
4424		error = NFSERR_SERVERFAULT;
4425		mbuf_freem(nd->nd_mreq);
4426		goto errout;
4427	}
4428
4429	/*
4430	 * Call newnfs_connect(), as required, and then newnfs_request().
4431	 */
4432	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4433	if (clp->lc_req.nr_client == NULL) {
4434		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4435			error = ECONNREFUSED;
4436			nfsrv_freesession(sep, NULL);
4437		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4438			error = newnfs_connect(NULL, &clp->lc_req, cred,
4439			    NULL, 1);
4440		else
4441			error = newnfs_connect(NULL, &clp->lc_req, cred,
4442			    NULL, 3);
4443	}
4444	newnfs_sndunlock(&clp->lc_req.nr_lock);
4445	if (!error) {
4446		if ((nd->nd_flag & ND_NFSV41) != 0) {
4447			KASSERT(sep != NULL, ("sep NULL"));
4448			if (sep->sess_cbsess.nfsess_xprt != NULL)
4449				error = newnfs_request(nd, NULL, clp,
4450				    &clp->lc_req, NULL, NULL, cred,
4451				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4452				    1, NULL, &sep->sess_cbsess);
4453			else {
4454				/*
4455				 * This should probably never occur, but if a
4456				 * client somehow does an RPC without a
4457				 * SequenceID Op that causes a callback just
4458				 * after the nfsd threads have been terminated
4459				 * and restared we could conceivably get here
4460				 * without a backchannel xprt.
4461				 */
4462				printf("nfsrv_docallback: no xprt\n");
4463				error = ECONNREFUSED;
4464			}
4465			nfsrv_freesession(sep, NULL);
4466		} else
4467			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4468			    NULL, NULL, cred, clp->lc_program,
4469			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4470	}
4471errout:
4472	NFSFREECRED(cred);
4473
4474	/*
4475	 * If error is set here, the Callback path isn't working
4476	 * properly, so twiddle the appropriate LCL_ flags.
4477	 * (nd_repstat != 0 indicates the Callback path is working,
4478	 *  but the callback failed on the client.)
4479	 */
4480	if (error) {
4481		/*
4482		 * Mark the callback pathway down, which disabled issuing
4483		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4484		 */
4485		NFSLOCKSTATE();
4486		clp->lc_flags |= LCL_CBDOWN;
4487		NFSUNLOCKSTATE();
4488	} else {
4489		/*
4490		 * Callback worked. If the callback path was down, disable
4491		 * callbacks, so no more delegations will be issued. (This
4492		 * is done on the assumption that the callback pathway is
4493		 * flakey.)
4494		 */
4495		NFSLOCKSTATE();
4496		if (clp->lc_flags & LCL_CBDOWN)
4497			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4498		NFSUNLOCKSTATE();
4499		if (nd->nd_repstat)
4500			error = nd->nd_repstat;
4501		else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4502			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4503			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4504			    p, NULL);
4505		mbuf_freem(nd->nd_mrep);
4506	}
4507	NFSLOCKSTATE();
4508	clp->lc_cbref--;
4509	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4510		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4511		wakeup(clp);
4512	}
4513	NFSUNLOCKSTATE();
4514
4515	NFSEXITCODE(error);
4516	return (error);
4517}
4518
4519/*
4520 * Set up the compound RPC for the callback.
4521 */
4522static int
4523nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4524    uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4525{
4526	uint32_t *tl;
4527	int error, len;
4528
4529	len = strlen(optag);
4530	(void)nfsm_strtom(nd, optag, len);
4531	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4532	if ((nd->nd_flag & ND_NFSV41) != 0) {
4533		*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4534		*tl++ = txdr_unsigned(callback);
4535		*tl++ = txdr_unsigned(2);
4536		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4537		error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4538		if (error != 0)
4539			return (error);
4540		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4541		*tl = txdr_unsigned(op);
4542	} else {
4543		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4544		*tl++ = txdr_unsigned(callback);
4545		*tl++ = txdr_unsigned(1);
4546		*tl = txdr_unsigned(op);
4547	}
4548	return (0);
4549}
4550
4551/*
4552 * Return the next index# for a clientid. Mostly just increment and return
4553 * the next one, but... if the 32bit unsigned does actually wrap around,
4554 * it should be rebooted.
4555 * At an average rate of one new client per second, it will wrap around in
4556 * approximately 136 years. (I think the server will have been shut
4557 * down or rebooted before then.)
4558 */
4559static u_int32_t
4560nfsrv_nextclientindex(void)
4561{
4562	static u_int32_t client_index = 0;
4563
4564	client_index++;
4565	if (client_index != 0)
4566		return (client_index);
4567
4568	printf("%s: out of clientids\n", __func__);
4569	return (client_index);
4570}
4571
4572/*
4573 * Return the next index# for a stateid. Mostly just increment and return
4574 * the next one, but... if the 32bit unsigned does actually wrap around
4575 * (will a BSD server stay up that long?), find
4576 * new start and end values.
4577 */
4578static u_int32_t
4579nfsrv_nextstateindex(struct nfsclient *clp)
4580{
4581	struct nfsstate *stp;
4582	int i;
4583	u_int32_t canuse, min_index, max_index;
4584
4585	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4586		clp->lc_stateindex++;
4587		if (clp->lc_stateindex != clp->lc_statemaxindex)
4588			return (clp->lc_stateindex);
4589	}
4590
4591	/*
4592	 * Yuck, we've hit the end.
4593	 * Look for a new min and max.
4594	 */
4595	min_index = 0;
4596	max_index = 0xffffffff;
4597	for (i = 0; i < nfsrv_statehashsize; i++) {
4598	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4599		if (stp->ls_stateid.other[2] > 0x80000000) {
4600		    if (stp->ls_stateid.other[2] < max_index)
4601			max_index = stp->ls_stateid.other[2];
4602		} else {
4603		    if (stp->ls_stateid.other[2] > min_index)
4604			min_index = stp->ls_stateid.other[2];
4605		}
4606	    }
4607	}
4608
4609	/*
4610	 * Yikes, highly unlikely, but I'll handle it anyhow.
4611	 */
4612	if (min_index == 0x80000000 && max_index == 0x80000001) {
4613	    canuse = 0;
4614	    /*
4615	     * Loop around until we find an unused entry. Return that
4616	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4617	     * (This is one of those rare cases where a goto is the
4618	     *  cleanest way to code the loop.)
4619	     */
4620tryagain:
4621	    for (i = 0; i < nfsrv_statehashsize; i++) {
4622		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4623		    if (stp->ls_stateid.other[2] == canuse) {
4624			canuse++;
4625			goto tryagain;
4626		    }
4627		}
4628	    }
4629	    clp->lc_flags |= LCL_INDEXNOTOK;
4630	    return (canuse);
4631	}
4632
4633	/*
4634	 * Ok to start again from min + 1.
4635	 */
4636	clp->lc_stateindex = min_index + 1;
4637	clp->lc_statemaxindex = max_index;
4638	clp->lc_flags &= ~LCL_INDEXNOTOK;
4639	return (clp->lc_stateindex);
4640}
4641
4642/*
4643 * The following functions handle the stable storage file that deals with
4644 * the edge conditions described in RFC3530 Sec. 8.6.3.
4645 * The file is as follows:
4646 * - a single record at the beginning that has the lease time of the
4647 *   previous server instance (before the last reboot) and the nfsrvboottime
4648 *   values for the previous server boots.
4649 *   These previous boot times are used to ensure that the current
4650 *   nfsrvboottime does not, somehow, get set to a previous one.
4651 *   (This is important so that Stale ClientIDs and StateIDs can
4652 *    be recognized.)
4653 *   The number of previous nfsvrboottime values preceeds the list.
4654 * - followed by some number of appended records with:
4655 *   - client id string
4656 *   - flag that indicates it is a record revoking state via lease
4657 *     expiration or similar
4658 *     OR has successfully acquired state.
4659 * These structures vary in length, with the client string at the end, up
4660 * to NFSV4_OPAQUELIMIT in size.
4661 *
4662 * At the end of the grace period, the file is truncated, the first
4663 * record is rewritten with updated information and any acquired state
4664 * records for successful reclaims of state are written.
4665 *
4666 * Subsequent records are appended when the first state is issued to
4667 * a client and when state is revoked for a client.
4668 *
4669 * When reading the file in, state issued records that come later in
4670 * the file override older ones, since the append log is in cronological order.
4671 * If, for some reason, the file can't be read, the grace period is
4672 * immediately terminated and all reclaims get NFSERR_NOGRACE.
4673 */
4674
4675/*
4676 * Read in the stable storage file. Called by nfssvc() before the nfsd
4677 * processes start servicing requests.
4678 */
4679APPLESTATIC void
4680nfsrv_setupstable(NFSPROC_T *p)
4681{
4682	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4683	struct nfsrv_stable *sp, *nsp;
4684	struct nfst_rec *tsp;
4685	int error, i, tryagain;
4686	off_t off = 0;
4687	ssize_t aresid, len;
4688
4689	/*
4690	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4691	 * a reboot, so state has not been lost.
4692	 */
4693	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4694		return;
4695	/*
4696	 * Set Grace over just until the file reads successfully.
4697	 */
4698	nfsrvboottime = time_second;
4699	LIST_INIT(&sf->nsf_head);
4700	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4701	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4702	if (sf->nsf_fp == NULL)
4703		return;
4704	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4705	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4706	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4707	if (error || aresid || sf->nsf_numboots == 0 ||
4708		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4709		return;
4710
4711	/*
4712	 * Now, read in the boottimes.
4713	 */
4714	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4715		sizeof (time_t), M_TEMP, M_WAITOK);
4716	off = sizeof (struct nfsf_rec);
4717	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4718	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4719	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4720	if (error || aresid) {
4721		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4722		sf->nsf_bootvals = NULL;
4723		return;
4724	}
4725
4726	/*
4727	 * Make sure this nfsrvboottime is different from all recorded
4728	 * previous ones.
4729	 */
4730	do {
4731		tryagain = 0;
4732		for (i = 0; i < sf->nsf_numboots; i++) {
4733			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4734				nfsrvboottime++;
4735				tryagain = 1;
4736				break;
4737			}
4738		}
4739	} while (tryagain);
4740
4741	sf->nsf_flags |= NFSNSF_OK;
4742	off += (sf->nsf_numboots * sizeof (time_t));
4743
4744	/*
4745	 * Read through the file, building a list of records for grace
4746	 * checking.
4747	 * Each record is between sizeof (struct nfst_rec) and
4748	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4749	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4750	 */
4751	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4752		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4753	do {
4754	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4755	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4756	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4757	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4758	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4759		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4760		/*
4761		 * Yuck, the file has been corrupted, so just return
4762		 * after clearing out any restart state, so the grace period
4763		 * is over.
4764		 */
4765		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4766			LIST_REMOVE(sp, nst_list);
4767			free((caddr_t)sp, M_TEMP);
4768		}
4769		free((caddr_t)tsp, M_TEMP);
4770		sf->nsf_flags &= ~NFSNSF_OK;
4771		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4772		sf->nsf_bootvals = NULL;
4773		return;
4774	    }
4775	    if (len > 0) {
4776		off += sizeof (struct nfst_rec) + tsp->len - 1;
4777		/*
4778		 * Search the list for a matching client.
4779		 */
4780		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4781			if (tsp->len == sp->nst_len &&
4782			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4783				break;
4784		}
4785		if (sp == LIST_END(&sf->nsf_head)) {
4786			sp = (struct nfsrv_stable *)malloc(tsp->len +
4787				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4788				M_WAITOK);
4789			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4790				sizeof (struct nfst_rec) + tsp->len - 1);
4791			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4792		} else {
4793			if (tsp->flag == NFSNST_REVOKE)
4794				sp->nst_flag |= NFSNST_REVOKE;
4795			else
4796				/*
4797				 * A subsequent timestamp indicates the client
4798				 * did a setclientid/confirm and any previous
4799				 * revoke is no longer relevant.
4800				 */
4801				sp->nst_flag &= ~NFSNST_REVOKE;
4802		}
4803	    }
4804	} while (len > 0);
4805	free((caddr_t)tsp, M_TEMP);
4806	sf->nsf_flags = NFSNSF_OK;
4807	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4808		NFSRV_LEASEDELTA;
4809}
4810
4811/*
4812 * Update the stable storage file, now that the grace period is over.
4813 */
4814APPLESTATIC void
4815nfsrv_updatestable(NFSPROC_T *p)
4816{
4817	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4818	struct nfsrv_stable *sp, *nsp;
4819	int i;
4820	struct nfsvattr nva;
4821	vnode_t vp;
4822#if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4823	mount_t mp = NULL;
4824#endif
4825	int error;
4826
4827	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4828		return;
4829	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4830	/*
4831	 * Ok, we need to rewrite the stable storage file.
4832	 * - truncate to 0 length
4833	 * - write the new first structure
4834	 * - loop through the data structures, writing out any that
4835	 *   have timestamps older than the old boot
4836	 */
4837	if (sf->nsf_bootvals) {
4838		sf->nsf_numboots++;
4839		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4840			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4841	} else {
4842		sf->nsf_numboots = 1;
4843		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4844			M_TEMP, M_WAITOK);
4845	}
4846	sf->nsf_bootvals[0] = nfsrvboottime;
4847	sf->nsf_lease = nfsrv_lease;
4848	NFSVNO_ATTRINIT(&nva);
4849	NFSVNO_SETATTRVAL(&nva, size, 0);
4850	vp = NFSFPVNODE(sf->nsf_fp);
4851	vn_start_write(vp, &mp, V_WAIT);
4852	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4853		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4854		    NULL);
4855		NFSVOPUNLOCK(vp, 0);
4856	} else
4857		error = EPERM;
4858	vn_finished_write(mp);
4859	if (!error)
4860	    error = NFSD_RDWR(UIO_WRITE, vp,
4861		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4862		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4863	if (!error)
4864	    error = NFSD_RDWR(UIO_WRITE, vp,
4865		(caddr_t)sf->nsf_bootvals,
4866		sf->nsf_numboots * sizeof (time_t),
4867		(off_t)(sizeof (struct nfsf_rec)),
4868		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4869	free((caddr_t)sf->nsf_bootvals, M_TEMP);
4870	sf->nsf_bootvals = NULL;
4871	if (error) {
4872		sf->nsf_flags &= ~NFSNSF_OK;
4873		printf("EEK! Can't write NfsV4 stable storage file\n");
4874		return;
4875	}
4876	sf->nsf_flags |= NFSNSF_OK;
4877
4878	/*
4879	 * Loop through the list and write out timestamp records for
4880	 * any clients that successfully reclaimed state.
4881	 */
4882	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4883		if (sp->nst_flag & NFSNST_GOTSTATE) {
4884			nfsrv_writestable(sp->nst_client, sp->nst_len,
4885				NFSNST_NEWSTATE, p);
4886			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4887		}
4888		LIST_REMOVE(sp, nst_list);
4889		free((caddr_t)sp, M_TEMP);
4890	}
4891	nfsrv_backupstable();
4892}
4893
4894/*
4895 * Append a record to the stable storage file.
4896 */
4897APPLESTATIC void
4898nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4899{
4900	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4901	struct nfst_rec *sp;
4902	int error;
4903
4904	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4905		return;
4906	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4907		len - 1, M_TEMP, M_WAITOK);
4908	sp->len = len;
4909	NFSBCOPY(client, sp->client, len);
4910	sp->flag = flag;
4911	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4912	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4913	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4914	free((caddr_t)sp, M_TEMP);
4915	if (error) {
4916		sf->nsf_flags &= ~NFSNSF_OK;
4917		printf("EEK! Can't write NfsV4 stable storage file\n");
4918	}
4919}
4920
4921/*
4922 * This function is called during the grace period to mark a client
4923 * that successfully reclaimed state.
4924 */
4925static void
4926nfsrv_markstable(struct nfsclient *clp)
4927{
4928	struct nfsrv_stable *sp;
4929
4930	/*
4931	 * First find the client structure.
4932	 */
4933	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4934		if (sp->nst_len == clp->lc_idlen &&
4935		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4936			break;
4937	}
4938	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4939		return;
4940
4941	/*
4942	 * Now, just mark it and set the nfsclient back pointer.
4943	 */
4944	sp->nst_flag |= NFSNST_GOTSTATE;
4945	sp->nst_clp = clp;
4946}
4947
4948/*
4949 * This function is called when a NFSv4.1 client does a ReclaimComplete.
4950 * Very similar to nfsrv_markstable(), except for the flag being set.
4951 */
4952static void
4953nfsrv_markreclaim(struct nfsclient *clp)
4954{
4955	struct nfsrv_stable *sp;
4956
4957	/*
4958	 * First find the client structure.
4959	 */
4960	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4961		if (sp->nst_len == clp->lc_idlen &&
4962		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4963			break;
4964	}
4965	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4966		return;
4967
4968	/*
4969	 * Now, just set the flag.
4970	 */
4971	sp->nst_flag |= NFSNST_RECLAIMED;
4972}
4973
4974/*
4975 * This function is called for a reclaim, to see if it gets grace.
4976 * It returns 0 if a reclaim is allowed, 1 otherwise.
4977 */
4978static int
4979nfsrv_checkstable(struct nfsclient *clp)
4980{
4981	struct nfsrv_stable *sp;
4982
4983	/*
4984	 * First, find the entry for the client.
4985	 */
4986	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4987		if (sp->nst_len == clp->lc_idlen &&
4988		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4989			break;
4990	}
4991
4992	/*
4993	 * If not in the list, state was revoked or no state was issued
4994	 * since the previous reboot, a reclaim is denied.
4995	 */
4996	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4997	    (sp->nst_flag & NFSNST_REVOKE) ||
4998	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4999		return (1);
5000	return (0);
5001}
5002
5003/*
5004 * Test for and try to clear out a conflicting client. This is called by
5005 * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5006 * a found.
5007 * The trick here is that it can't revoke a conflicting client with an
5008 * expired lease unless it holds the v4root lock, so...
5009 * If no v4root lock, get the lock and return 1 to indicate "try again".
5010 * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5011 * the revocation worked and the conflicting client is "bye, bye", so it
5012 * can be tried again.
5013 * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
5014 * Unlocks State before a non-zero value is returned.
5015 */
5016static int
5017nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5018    NFSPROC_T *p)
5019{
5020	int gotlock, lktype = 0;
5021
5022	/*
5023	 * If lease hasn't expired, we can't fix it.
5024	 */
5025	if (clp->lc_expiry >= NFSD_MONOSEC ||
5026	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
5027		return (0);
5028	if (*haslockp == 0) {
5029		NFSUNLOCKSTATE();
5030		if (vp != NULL) {
5031			lktype = NFSVOPISLOCKED(vp);
5032			NFSVOPUNLOCK(vp, 0);
5033		}
5034		NFSLOCKV4ROOTMUTEX();
5035		nfsv4_relref(&nfsv4rootfs_lock);
5036		do {
5037			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5038			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5039		} while (!gotlock);
5040		NFSUNLOCKV4ROOTMUTEX();
5041		*haslockp = 1;
5042		if (vp != NULL) {
5043			NFSVOPLOCK(vp, lktype | LK_RETRY);
5044			if ((vp->v_iflag & VI_DOOMED) != 0)
5045				return (2);
5046		}
5047		return (1);
5048	}
5049	NFSUNLOCKSTATE();
5050
5051	/*
5052	 * Ok, we can expire the conflicting client.
5053	 */
5054	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5055	nfsrv_backupstable();
5056	nfsrv_cleanclient(clp, p);
5057	nfsrv_freedeleglist(&clp->lc_deleg);
5058	nfsrv_freedeleglist(&clp->lc_olddeleg);
5059	LIST_REMOVE(clp, lc_hash);
5060	nfsrv_zapclient(clp, p);
5061	return (1);
5062}
5063
5064/*
5065 * Resolve a delegation conflict.
5066 * Returns 0 to indicate the conflict was resolved without sleeping.
5067 * Return -1 to indicate that the caller should check for conflicts again.
5068 * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5069 *
5070 * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5071 * for a return of 0, since there was no sleep and it could be required
5072 * later. It is released for a return of NFSERR_DELAY, since the caller
5073 * will return that error. It is released when a sleep was done waiting
5074 * for the delegation to be returned or expire (so that other nfsds can
5075 * handle ops). Then, it must be acquired for the write to stable storage.
5076 * (This function is somewhat similar to nfsrv_clientconflict(), but
5077 *  the semantics differ in a couple of subtle ways. The return of 0
5078 *  indicates the conflict was resolved without sleeping here, not
5079 *  that the conflict can't be resolved and the handling of nfsv4root_lock
5080 *  differs, as noted above.)
5081 * Unlocks State before returning a non-zero value.
5082 */
5083static int
5084nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5085    vnode_t vp)
5086{
5087	struct nfsclient *clp = stp->ls_clp;
5088	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5089	nfsv4stateid_t tstateid;
5090	fhandle_t tfh;
5091
5092	/*
5093	 * If the conflict is with an old delegation...
5094	 */
5095	if (stp->ls_flags & NFSLCK_OLDDELEG) {
5096		/*
5097		 * You can delete it, if it has expired.
5098		 */
5099		if (clp->lc_delegtime < NFSD_MONOSEC) {
5100			nfsrv_freedeleg(stp);
5101			NFSUNLOCKSTATE();
5102			error = -1;
5103			goto out;
5104		}
5105		NFSUNLOCKSTATE();
5106		/*
5107		 * During this delay, the old delegation could expire or it
5108		 * could be recovered by the client via an Open with
5109		 * CLAIM_DELEGATE_PREV.
5110		 * Release the nfsv4root_lock, if held.
5111		 */
5112		if (*haslockp) {
5113			*haslockp = 0;
5114			NFSLOCKV4ROOTMUTEX();
5115			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5116			NFSUNLOCKV4ROOTMUTEX();
5117		}
5118		error = NFSERR_DELAY;
5119		goto out;
5120	}
5121
5122	/*
5123	 * It's a current delegation, so:
5124	 * - check to see if the delegation has expired
5125	 *   - if so, get the v4root lock and then expire it
5126	 */
5127	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
5128		/*
5129		 * - do a recall callback, since not yet done
5130		 * For now, never allow truncate to be set. To use
5131		 * truncate safely, it must be guaranteed that the
5132		 * Remove, Rename or Setattr with size of 0 will
5133		 * succeed and that would require major changes to
5134		 * the VFS/Vnode OPs.
5135		 * Set the expiry time large enough so that it won't expire
5136		 * until after the callback, then set it correctly, once
5137		 * the callback is done. (The delegation will now time
5138		 * out whether or not the Recall worked ok. The timeout
5139		 * will be extended when ops are done on the delegation
5140		 * stateid, up to the timelimit.)
5141		 */
5142		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5143		    NFSRV_LEASEDELTA;
5144		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
5145		    NFSRV_LEASEDELTA;
5146		stp->ls_flags |= NFSLCK_DELEGRECALL;
5147
5148		/*
5149		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5150		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5151		 * in order to try and avoid a race that could happen
5152		 * when a CBRecall request passed the Open reply with
5153		 * the delegation in it when transitting the network.
5154		 * Since nfsrv_docallback will sleep, don't use stp after
5155		 * the call.
5156		 */
5157		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5158		    sizeof (tstateid));
5159		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5160		    sizeof (tfh));
5161		NFSUNLOCKSTATE();
5162		if (*haslockp) {
5163			*haslockp = 0;
5164			NFSLOCKV4ROOTMUTEX();
5165			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5166			NFSUNLOCKV4ROOTMUTEX();
5167		}
5168		retrycnt = 0;
5169		do {
5170		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5171			&tstateid, 0, &tfh, NULL, NULL, p);
5172		    retrycnt++;
5173		} while ((error == NFSERR_BADSTATEID ||
5174		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5175		error = NFSERR_DELAY;
5176		goto out;
5177	}
5178
5179	if (clp->lc_expiry >= NFSD_MONOSEC &&
5180	    stp->ls_delegtime >= NFSD_MONOSEC) {
5181		NFSUNLOCKSTATE();
5182		/*
5183		 * A recall has been done, but it has not yet expired.
5184		 * So, RETURN_DELAY.
5185		 */
5186		if (*haslockp) {
5187			*haslockp = 0;
5188			NFSLOCKV4ROOTMUTEX();
5189			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5190			NFSUNLOCKV4ROOTMUTEX();
5191		}
5192		error = NFSERR_DELAY;
5193		goto out;
5194	}
5195
5196	/*
5197	 * If we don't yet have the lock, just get it and then return,
5198	 * since we need that before deleting expired state, such as
5199	 * this delegation.
5200	 * When getting the lock, unlock the vnode, so other nfsds that
5201	 * are in progress, won't get stuck waiting for the vnode lock.
5202	 */
5203	if (*haslockp == 0) {
5204		NFSUNLOCKSTATE();
5205		if (vp != NULL) {
5206			lktype = NFSVOPISLOCKED(vp);
5207			NFSVOPUNLOCK(vp, 0);
5208		}
5209		NFSLOCKV4ROOTMUTEX();
5210		nfsv4_relref(&nfsv4rootfs_lock);
5211		do {
5212			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5213			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5214		} while (!gotlock);
5215		NFSUNLOCKV4ROOTMUTEX();
5216		*haslockp = 1;
5217		if (vp != NULL) {
5218			NFSVOPLOCK(vp, lktype | LK_RETRY);
5219			if ((vp->v_iflag & VI_DOOMED) != 0) {
5220				*haslockp = 0;
5221				NFSLOCKV4ROOTMUTEX();
5222				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5223				NFSUNLOCKV4ROOTMUTEX();
5224				error = NFSERR_PERM;
5225				goto out;
5226			}
5227		}
5228		error = -1;
5229		goto out;
5230	}
5231
5232	NFSUNLOCKSTATE();
5233	/*
5234	 * Ok, we can delete the expired delegation.
5235	 * First, write the Revoke record to stable storage and then
5236	 * clear out the conflict.
5237	 * Since all other nfsd threads are now blocked, we can safely
5238	 * sleep without the state changing.
5239	 */
5240	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5241	nfsrv_backupstable();
5242	if (clp->lc_expiry < NFSD_MONOSEC) {
5243		nfsrv_cleanclient(clp, p);
5244		nfsrv_freedeleglist(&clp->lc_deleg);
5245		nfsrv_freedeleglist(&clp->lc_olddeleg);
5246		LIST_REMOVE(clp, lc_hash);
5247		zapped_clp = 1;
5248	} else {
5249		nfsrv_freedeleg(stp);
5250		zapped_clp = 0;
5251	}
5252	if (zapped_clp)
5253		nfsrv_zapclient(clp, p);
5254	error = -1;
5255
5256out:
5257	NFSEXITCODE(error);
5258	return (error);
5259}
5260
5261/*
5262 * Check for a remove allowed, if remove is set to 1 and get rid of
5263 * delegations.
5264 */
5265APPLESTATIC int
5266nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
5267{
5268	struct nfsstate *stp;
5269	struct nfslockfile *lfp;
5270	int error, haslock = 0;
5271	fhandle_t nfh;
5272
5273	/*
5274	 * First, get the lock file structure.
5275	 * (A return of -1 means no associated state, so remove ok.)
5276	 */
5277	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5278tryagain:
5279	NFSLOCKSTATE();
5280	if (!error)
5281		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5282	if (error) {
5283		NFSUNLOCKSTATE();
5284		if (haslock) {
5285			NFSLOCKV4ROOTMUTEX();
5286			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5287			NFSUNLOCKV4ROOTMUTEX();
5288		}
5289		if (error == -1)
5290			error = 0;
5291		goto out;
5292	}
5293
5294	/*
5295	 * Now, we must Recall any delegations.
5296	 */
5297	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
5298	if (error) {
5299		/*
5300		 * nfsrv_cleandeleg() unlocks state for non-zero
5301		 * return.
5302		 */
5303		if (error == -1)
5304			goto tryagain;
5305		if (haslock) {
5306			NFSLOCKV4ROOTMUTEX();
5307			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5308			NFSUNLOCKV4ROOTMUTEX();
5309		}
5310		goto out;
5311	}
5312
5313	/*
5314	 * Now, look for a conflicting open share.
5315	 */
5316	if (remove) {
5317		LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5318			if (stp->ls_flags & NFSLCK_WRITEDENY) {
5319				error = NFSERR_FILEOPEN;
5320				break;
5321			}
5322		}
5323	}
5324
5325	NFSUNLOCKSTATE();
5326	if (haslock) {
5327		NFSLOCKV4ROOTMUTEX();
5328		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5329		NFSUNLOCKV4ROOTMUTEX();
5330	}
5331
5332out:
5333	NFSEXITCODE(error);
5334	return (error);
5335}
5336
5337/*
5338 * Clear out all delegations for the file referred to by lfp.
5339 * May return NFSERR_DELAY, if there will be a delay waiting for
5340 * delegations to expire.
5341 * Returns -1 to indicate it slept while recalling a delegation.
5342 * This function has the side effect of deleting the nfslockfile structure,
5343 * if it no longer has associated state and didn't have to sleep.
5344 * Unlocks State before a non-zero value is returned.
5345 */
5346static int
5347nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5348    struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5349{
5350	struct nfsstate *stp, *nstp;
5351	int ret = 0;
5352
5353	stp = LIST_FIRST(&lfp->lf_deleg);
5354	while (stp != LIST_END(&lfp->lf_deleg)) {
5355		nstp = LIST_NEXT(stp, ls_file);
5356		if (stp->ls_clp != clp) {
5357			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5358			if (ret) {
5359				/*
5360				 * nfsrv_delegconflict() unlocks state
5361				 * when it returns non-zero.
5362				 */
5363				goto out;
5364			}
5365		}
5366		stp = nstp;
5367	}
5368out:
5369	NFSEXITCODE(ret);
5370	return (ret);
5371}
5372
5373/*
5374 * There are certain operations that, when being done outside of NFSv4,
5375 * require that any NFSv4 delegation for the file be recalled.
5376 * This function is to be called for those cases:
5377 * VOP_RENAME() - When a delegation is being recalled for any reason,
5378 *	the client may have to do Opens against the server, using the file's
5379 *	final component name. If the file has been renamed on the server,
5380 *	that component name will be incorrect and the Open will fail.
5381 * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5382 *	been removed on the server, if there is a delegation issued to
5383 *	that client for the file. I say "theoretically" since clients
5384 *	normally do an Access Op before the Open and that Access Op will
5385 *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5386 *	they will detect the file's removal in the same manner. (There is
5387 *	one case where RFC3530 allows a client to do an Open without first
5388 *	doing an Access Op, which is passage of a check against the ACE
5389 *	returned with a Write delegation, but current practice is to ignore
5390 *	the ACE and always do an Access Op.)
5391 *	Since the functions can only be called with an unlocked vnode, this
5392 *	can't be done at this time.
5393 * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5394 *	locks locally in the client, which are not visible to the server. To
5395 *	deal with this, issuing of delegations for a vnode must be disabled
5396 *	and all delegations for the vnode recalled. This is done via the
5397 *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5398 */
5399APPLESTATIC void
5400nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5401{
5402	time_t starttime;
5403	int error;
5404
5405	/*
5406	 * First, check to see if the server is currently running and it has
5407	 * been called for a regular file when issuing delegations.
5408	 */
5409	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5410	    nfsrv_issuedelegs == 0)
5411		return;
5412
5413	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5414	/*
5415	 * First, get a reference on the nfsv4rootfs_lock so that an
5416	 * exclusive lock cannot be acquired by another thread.
5417	 */
5418	NFSLOCKV4ROOTMUTEX();
5419	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5420	NFSUNLOCKV4ROOTMUTEX();
5421
5422	/*
5423	 * Now, call nfsrv_checkremove() in a loop while it returns
5424	 * NFSERR_DELAY. Return upon any other error or when timed out.
5425	 */
5426	starttime = NFSD_MONOSEC;
5427	do {
5428		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5429			error = nfsrv_checkremove(vp, 0, p);
5430			NFSVOPUNLOCK(vp, 0);
5431		} else
5432			error = EPERM;
5433		if (error == NFSERR_DELAY) {
5434			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5435				break;
5436			/* Sleep for a short period of time */
5437			(void) nfs_catnap(PZERO, 0, "nfsremove");
5438		}
5439	} while (error == NFSERR_DELAY);
5440	NFSLOCKV4ROOTMUTEX();
5441	nfsv4_relref(&nfsv4rootfs_lock);
5442	NFSUNLOCKV4ROOTMUTEX();
5443}
5444
5445APPLESTATIC void
5446nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5447{
5448
5449#ifdef VV_DISABLEDELEG
5450	/*
5451	 * First, flag issuance of delegations disabled.
5452	 */
5453	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5454#endif
5455
5456	/*
5457	 * Then call nfsd_recalldelegation() to get rid of all extant
5458	 * delegations.
5459	 */
5460	nfsd_recalldelegation(vp, p);
5461}
5462
5463/*
5464 * Check for conflicting locks, etc. and then get rid of delegations.
5465 * (At one point I thought that I should get rid of delegations for any
5466 *  Setattr, since it could potentially disallow the I/O op (read or write)
5467 *  allowed by the delegation. However, Setattr Ops that aren't changing
5468 *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5469 *  for the same client or a different one, so I decided to only get rid
5470 *  of delegations for other clients when the size is being changed.)
5471 * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5472 * as Write backs, even if there is no delegation, so it really isn't any
5473 * different?)
5474 */
5475APPLESTATIC int
5476nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5477    nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5478    struct nfsexstuff *exp, NFSPROC_T *p)
5479{
5480	struct nfsstate st, *stp = &st;
5481	struct nfslock lo, *lop = &lo;
5482	int error = 0;
5483	nfsquad_t clientid;
5484
5485	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5486		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5487		lop->lo_first = nvap->na_size;
5488	} else {
5489		stp->ls_flags = 0;
5490		lop->lo_first = 0;
5491	}
5492	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5493	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5494	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5495	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5496		stp->ls_flags |= NFSLCK_SETATTR;
5497	if (stp->ls_flags == 0)
5498		goto out;
5499	lop->lo_end = NFS64BITSSET;
5500	lop->lo_flags = NFSLCK_WRITE;
5501	stp->ls_ownerlen = 0;
5502	stp->ls_op = NULL;
5503	stp->ls_uid = nd->nd_cred->cr_uid;
5504	stp->ls_stateid.seqid = stateidp->seqid;
5505	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5506	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5507	stp->ls_stateid.other[2] = stateidp->other[2];
5508	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5509	    stateidp, exp, nd, p);
5510
5511out:
5512	NFSEXITCODE2(error, nd);
5513	return (error);
5514}
5515
5516/*
5517 * Check for a write delegation and do a CBGETATTR if there is one, updating
5518 * the attributes, as required.
5519 * Should I return an error if I can't get the attributes? (For now, I'll
5520 * just return ok.
5521 */
5522APPLESTATIC int
5523nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5524    struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
5525    NFSPROC_T *p)
5526{
5527	struct nfsstate *stp;
5528	struct nfslockfile *lfp;
5529	struct nfsclient *clp;
5530	struct nfsvattr nva;
5531	fhandle_t nfh;
5532	int error = 0;
5533	nfsattrbit_t cbbits;
5534	u_quad_t delegfilerev;
5535
5536	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5537	if (!NFSNONZERO_ATTRBIT(&cbbits))
5538		goto out;
5539	if (nfsrv_writedelegcnt == 0)
5540		goto out;
5541
5542	/*
5543	 * Get the lock file structure.
5544	 * (A return of -1 means no associated state, so return ok.)
5545	 */
5546	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5547	NFSLOCKSTATE();
5548	if (!error)
5549		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5550	if (error) {
5551		NFSUNLOCKSTATE();
5552		if (error == -1)
5553			error = 0;
5554		goto out;
5555	}
5556
5557	/*
5558	 * Now, look for a write delegation.
5559	 */
5560	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5561		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5562			break;
5563	}
5564	if (stp == LIST_END(&lfp->lf_deleg)) {
5565		NFSUNLOCKSTATE();
5566		goto out;
5567	}
5568	clp = stp->ls_clp;
5569	delegfilerev = stp->ls_filerev;
5570
5571	/*
5572	 * If the Write delegation was issued as a part of this Compound RPC
5573	 * or if we have an Implied Clientid (used in a previous Op in this
5574	 * compound) and it is the client the delegation was issued to,
5575	 * just return ok.
5576	 * I also assume that it is from the same client iff the network
5577	 * host IP address is the same as the callback address. (Not
5578	 * exactly correct by the RFC, but avoids a lot of Getattr
5579	 * callbacks.)
5580	 */
5581	if (nd->nd_compref == stp->ls_compref ||
5582	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5583	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5584	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5585		NFSUNLOCKSTATE();
5586		goto out;
5587	}
5588
5589	/*
5590	 * We are now done with the delegation state structure,
5591	 * so the statelock can be released and we can now tsleep().
5592	 */
5593
5594	/*
5595	 * Now, we must do the CB Getattr callback, to see if Change or Size
5596	 * has changed.
5597	 */
5598	if (clp->lc_expiry >= NFSD_MONOSEC) {
5599		NFSUNLOCKSTATE();
5600		NFSVNO_ATTRINIT(&nva);
5601		nva.na_filerev = NFS64BITSSET;
5602		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5603		    0, &nfh, &nva, &cbbits, p);
5604		if (!error) {
5605			if ((nva.na_filerev != NFS64BITSSET &&
5606			    nva.na_filerev > delegfilerev) ||
5607			    (NFSVNO_ISSETSIZE(&nva) &&
5608			     nva.na_size != nvap->na_size)) {
5609				error = nfsvno_updfilerev(vp, nvap, cred, p);
5610				if (NFSVNO_ISSETSIZE(&nva))
5611					nvap->na_size = nva.na_size;
5612			}
5613		} else
5614			error = 0;	/* Ignore callback errors for now. */
5615	} else {
5616		NFSUNLOCKSTATE();
5617	}
5618
5619out:
5620	NFSEXITCODE2(error, nd);
5621	return (error);
5622}
5623
5624/*
5625 * This function looks for openowners that haven't had any opens for
5626 * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5627 * is set.
5628 */
5629APPLESTATIC void
5630nfsrv_throwawayopens(NFSPROC_T *p)
5631{
5632	struct nfsclient *clp, *nclp;
5633	struct nfsstate *stp, *nstp;
5634	int i;
5635
5636	NFSLOCKSTATE();
5637	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5638	/*
5639	 * For each client...
5640	 */
5641	for (i = 0; i < nfsrv_clienthashsize; i++) {
5642	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5643		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5644			if (LIST_EMPTY(&stp->ls_open) &&
5645			    (stp->ls_noopens > NFSNOOPEN ||
5646			     (nfsrv_openpluslock * 2) >
5647			     nfsrv_v4statelimit))
5648				nfsrv_freeopenowner(stp, 0, p);
5649		}
5650	    }
5651	}
5652	NFSUNLOCKSTATE();
5653}
5654
5655/*
5656 * This function checks to see if the credentials are the same.
5657 * Returns 1 for not same, 0 otherwise.
5658 */
5659static int
5660nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5661{
5662
5663	if (nd->nd_flag & ND_GSS) {
5664		if (!(clp->lc_flags & LCL_GSS))
5665			return (1);
5666		if (clp->lc_flags & LCL_NAME) {
5667			if (nd->nd_princlen != clp->lc_namelen ||
5668			    NFSBCMP(nd->nd_principal, clp->lc_name,
5669				clp->lc_namelen))
5670				return (1);
5671			else
5672				return (0);
5673		}
5674		if (nd->nd_cred->cr_uid == clp->lc_uid)
5675			return (0);
5676		else
5677			return (1);
5678	} else if (clp->lc_flags & LCL_GSS)
5679		return (1);
5680	/*
5681	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5682	 * in RFC3530, which talks about principals, but doesn't say anything
5683	 * about uids for AUTH_SYS.)
5684	 */
5685	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5686		return (0);
5687	else
5688		return (1);
5689}
5690
5691/*
5692 * Calculate the lease expiry time.
5693 */
5694static time_t
5695nfsrv_leaseexpiry(void)
5696{
5697
5698	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5699		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5700	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5701}
5702
5703/*
5704 * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5705 */
5706static void
5707nfsrv_delaydelegtimeout(struct nfsstate *stp)
5708{
5709
5710	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5711		return;
5712
5713	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5714	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5715		stp->ls_delegtime += nfsrv_lease;
5716		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5717			stp->ls_delegtime = stp->ls_delegtimelimit;
5718	}
5719}
5720
5721/*
5722 * This function checks to see if there is any other state associated
5723 * with the openowner for this Open.
5724 * It returns 1 if there is no other state, 0 otherwise.
5725 */
5726static int
5727nfsrv_nootherstate(struct nfsstate *stp)
5728{
5729	struct nfsstate *tstp;
5730
5731	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5732		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5733			return (0);
5734	}
5735	return (1);
5736}
5737
5738/*
5739 * Create a list of lock deltas (changes to local byte range locking
5740 * that can be rolled back using the list) and apply the changes via
5741 * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5742 * the rollback or update function will be called after this.
5743 * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5744 * call fails. If it returns an error, it will unlock the list.
5745 */
5746static int
5747nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5748    uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5749{
5750	struct nfslock *lop, *nlop;
5751	int error = 0;
5752
5753	/* Loop through the list of locks. */
5754	lop = LIST_FIRST(&lfp->lf_locallock);
5755	while (first < end && lop != NULL) {
5756		nlop = LIST_NEXT(lop, lo_lckowner);
5757		if (first >= lop->lo_end) {
5758			/* not there yet */
5759			lop = nlop;
5760		} else if (first < lop->lo_first) {
5761			/* new one starts before entry in list */
5762			if (end <= lop->lo_first) {
5763				/* no overlap between old and new */
5764				error = nfsrv_dolocal(vp, lfp, flags,
5765				    NFSLCK_UNLOCK, first, end, cfp, p);
5766				if (error != 0)
5767					break;
5768				first = end;
5769			} else {
5770				/* handle fragment overlapped with new one */
5771				error = nfsrv_dolocal(vp, lfp, flags,
5772				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5773				    p);
5774				if (error != 0)
5775					break;
5776				first = lop->lo_first;
5777			}
5778		} else {
5779			/* new one overlaps this entry in list */
5780			if (end <= lop->lo_end) {
5781				/* overlaps all of new one */
5782				error = nfsrv_dolocal(vp, lfp, flags,
5783				    lop->lo_flags, first, end, cfp, p);
5784				if (error != 0)
5785					break;
5786				first = end;
5787			} else {
5788				/* handle fragment overlapped with new one */
5789				error = nfsrv_dolocal(vp, lfp, flags,
5790				    lop->lo_flags, first, lop->lo_end, cfp, p);
5791				if (error != 0)
5792					break;
5793				first = lop->lo_end;
5794				lop = nlop;
5795			}
5796		}
5797	}
5798	if (first < end && error == 0)
5799		/* handle fragment past end of list */
5800		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5801		    end, cfp, p);
5802
5803	NFSEXITCODE(error);
5804	return (error);
5805}
5806
5807/*
5808 * Local lock unlock. Unlock all byte ranges that are no longer locked
5809 * by NFSv4. To do this, unlock any subranges of first-->end that
5810 * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5811 * list. This list has all locks for the file held by other
5812 * <clientid, lockowner> tuples. The list is ordered by increasing
5813 * lo_first value, but may have entries that overlap each other, for
5814 * the case of read locks.
5815 */
5816static void
5817nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5818    uint64_t init_end, NFSPROC_T *p)
5819{
5820	struct nfslock *lop;
5821	uint64_t first, end, prevfirst;
5822
5823	first = init_first;
5824	end = init_end;
5825	while (first < init_end) {
5826		/* Loop through all nfs locks, adjusting first and end */
5827		prevfirst = 0;
5828		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5829			KASSERT(prevfirst <= lop->lo_first,
5830			    ("nfsv4 locks out of order"));
5831			KASSERT(lop->lo_first < lop->lo_end,
5832			    ("nfsv4 bogus lock"));
5833			prevfirst = lop->lo_first;
5834			if (first >= lop->lo_first &&
5835			    first < lop->lo_end)
5836				/*
5837				 * Overlaps with initial part, so trim
5838				 * off that initial part by moving first past
5839				 * it.
5840				 */
5841				first = lop->lo_end;
5842			else if (end > lop->lo_first &&
5843			    lop->lo_first > first) {
5844				/*
5845				 * This lock defines the end of the
5846				 * segment to unlock, so set end to the
5847				 * start of it and break out of the loop.
5848				 */
5849				end = lop->lo_first;
5850				break;
5851			}
5852			if (first >= end)
5853				/*
5854				 * There is no segment left to do, so
5855				 * break out of this loop and then exit
5856				 * the outer while() since first will be set
5857				 * to end, which must equal init_end here.
5858				 */
5859				break;
5860		}
5861		if (first < end) {
5862			/* Unlock this segment */
5863			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5864			    NFSLCK_READ, first, end, NULL, p);
5865			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5866			    first, end);
5867		}
5868		/*
5869		 * Now move past this segment and look for any further
5870		 * segment in the range, if there is one.
5871		 */
5872		first = end;
5873		end = init_end;
5874	}
5875}
5876
5877/*
5878 * Do the local lock operation and update the rollback list, as required.
5879 * Perform the rollback and return the error if nfsvno_advlock() fails.
5880 */
5881static int
5882nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5883    uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5884{
5885	struct nfsrollback *rlp;
5886	int error = 0, ltype, oldltype;
5887
5888	if (flags & NFSLCK_WRITE)
5889		ltype = F_WRLCK;
5890	else if (flags & NFSLCK_READ)
5891		ltype = F_RDLCK;
5892	else
5893		ltype = F_UNLCK;
5894	if (oldflags & NFSLCK_WRITE)
5895		oldltype = F_WRLCK;
5896	else if (oldflags & NFSLCK_READ)
5897		oldltype = F_RDLCK;
5898	else
5899		oldltype = F_UNLCK;
5900	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5901		/* nothing to do */
5902		goto out;
5903	error = nfsvno_advlock(vp, ltype, first, end, p);
5904	if (error != 0) {
5905		if (cfp != NULL) {
5906			cfp->cl_clientid.lval[0] = 0;
5907			cfp->cl_clientid.lval[1] = 0;
5908			cfp->cl_first = 0;
5909			cfp->cl_end = NFS64BITSSET;
5910			cfp->cl_flags = NFSLCK_WRITE;
5911			cfp->cl_ownerlen = 5;
5912			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5913		}
5914		nfsrv_locallock_rollback(vp, lfp, p);
5915	} else if (ltype != F_UNLCK) {
5916		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5917		    M_WAITOK);
5918		rlp->rlck_first = first;
5919		rlp->rlck_end = end;
5920		rlp->rlck_type = oldltype;
5921		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5922	}
5923
5924out:
5925	NFSEXITCODE(error);
5926	return (error);
5927}
5928
5929/*
5930 * Roll back local lock changes and free up the rollback list.
5931 */
5932static void
5933nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5934{
5935	struct nfsrollback *rlp, *nrlp;
5936
5937	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5938		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5939		    rlp->rlck_end, p);
5940		free(rlp, M_NFSDROLLBACK);
5941	}
5942	LIST_INIT(&lfp->lf_rollback);
5943}
5944
5945/*
5946 * Update local lock list and delete rollback list (ie now committed to the
5947 * local locks). Most of the work is done by the internal function.
5948 */
5949static void
5950nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5951    uint64_t end)
5952{
5953	struct nfsrollback *rlp, *nrlp;
5954	struct nfslock *new_lop, *other_lop;
5955
5956	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5957	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5958		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5959		    M_WAITOK);
5960	else
5961		other_lop = NULL;
5962	new_lop->lo_flags = flags;
5963	new_lop->lo_first = first;
5964	new_lop->lo_end = end;
5965	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5966	if (new_lop != NULL)
5967		free(new_lop, M_NFSDLOCK);
5968	if (other_lop != NULL)
5969		free(other_lop, M_NFSDLOCK);
5970
5971	/* and get rid of the rollback list */
5972	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5973		free(rlp, M_NFSDROLLBACK);
5974	LIST_INIT(&lfp->lf_rollback);
5975}
5976
5977/*
5978 * Lock the struct nfslockfile for local lock updating.
5979 */
5980static void
5981nfsrv_locklf(struct nfslockfile *lfp)
5982{
5983	int gotlock;
5984
5985	/* lf_usecount ensures *lfp won't be free'd */
5986	lfp->lf_usecount++;
5987	do {
5988		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5989		    NFSSTATEMUTEXPTR, NULL);
5990	} while (gotlock == 0);
5991	lfp->lf_usecount--;
5992}
5993
5994/*
5995 * Unlock the struct nfslockfile after local lock updating.
5996 */
5997static void
5998nfsrv_unlocklf(struct nfslockfile *lfp)
5999{
6000
6001	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6002}
6003
6004/*
6005 * Clear out all state for the NFSv4 server.
6006 * Must be called by a thread that can sleep when no nfsds are running.
6007 */
6008void
6009nfsrv_throwawayallstate(NFSPROC_T *p)
6010{
6011	struct nfsclient *clp, *nclp;
6012	struct nfslockfile *lfp, *nlfp;
6013	int i;
6014
6015	/*
6016	 * For each client, clean out the state and then free the structure.
6017	 */
6018	for (i = 0; i < nfsrv_clienthashsize; i++) {
6019		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
6020			nfsrv_cleanclient(clp, p);
6021			nfsrv_freedeleglist(&clp->lc_deleg);
6022			nfsrv_freedeleglist(&clp->lc_olddeleg);
6023			free(clp->lc_stateid, M_NFSDCLIENT);
6024			free(clp, M_NFSDCLIENT);
6025		}
6026	}
6027
6028	/*
6029	 * Also, free up any remaining lock file structures.
6030	 */
6031	for (i = 0; i < nfsrv_lockhashsize; i++) {
6032		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
6033			printf("nfsd unload: fnd a lock file struct\n");
6034			nfsrv_freenfslockfile(lfp);
6035		}
6036	}
6037}
6038
6039/*
6040 * Check the sequence# for the session and slot provided as an argument.
6041 * Also, renew the lease if the session will return NFS_OK.
6042 */
6043int
6044nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6045    uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6046    uint32_t *sflagsp, NFSPROC_T *p)
6047{
6048	struct nfsdsession *sep;
6049	struct nfssessionhash *shp;
6050	int error;
6051	SVCXPRT *savxprt;
6052
6053	shp = NFSSESSIONHASH(nd->nd_sessionid);
6054	NFSLOCKSESSION(shp);
6055	sep = nfsrv_findsession(nd->nd_sessionid);
6056	if (sep == NULL) {
6057		NFSUNLOCKSESSION(shp);
6058		return (NFSERR_BADSESSION);
6059	}
6060	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6061	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6062	if (error != 0) {
6063		NFSUNLOCKSESSION(shp);
6064		return (error);
6065	}
6066	if (cache_this != 0)
6067		nd->nd_flag |= ND_SAVEREPLY;
6068	/* Renew the lease. */
6069	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6070	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6071	nd->nd_flag |= ND_IMPLIEDCLID;
6072
6073	/*
6074	 * If this session handles the backchannel, save the nd_xprt for this
6075	 * RPC, since this is the one being used.
6076	 * RFC-5661 specifies that the fore channel will be implicitly
6077	 * bound by a Sequence operation.  However, since some NFSv4.1 clients
6078	 * erroneously assumed that the back channel would be implicitly
6079	 * bound as well, do the implicit binding unless a
6080	 * BindConnectiontoSession has already been done on the session.
6081	 */
6082	if (sep->sess_clp->lc_req.nr_client != NULL &&
6083	    sep->sess_cbsess.nfsess_xprt != nd->nd_xprt &&
6084	    (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0 &&
6085	    (sep->sess_clp->lc_flags & LCL_DONEBINDCONN) == 0) {
6086		NFSD_DEBUG(2,
6087		    "nfsrv_checksequence: implicit back channel bind\n");
6088		savxprt = sep->sess_cbsess.nfsess_xprt;
6089		SVC_ACQUIRE(nd->nd_xprt);
6090		nd->nd_xprt->xp_p2 =
6091		    sep->sess_clp->lc_req.nr_client->cl_private;
6092		nd->nd_xprt->xp_idletimeout = 0;	/* Disable timeout. */
6093		sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6094		if (savxprt != NULL)
6095			SVC_RELEASE(savxprt);
6096	}
6097
6098	*sflagsp = 0;
6099	if (sep->sess_clp->lc_req.nr_client == NULL)
6100		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
6101	NFSUNLOCKSESSION(shp);
6102	if (error == NFSERR_EXPIRED) {
6103		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6104		error = 0;
6105	} else if (error == NFSERR_ADMINREVOKED) {
6106		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6107		error = 0;
6108	}
6109	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6110	return (0);
6111}
6112
6113/*
6114 * Check/set reclaim complete for this session/clientid.
6115 */
6116int
6117nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6118{
6119	struct nfsdsession *sep;
6120	struct nfssessionhash *shp;
6121	int error = 0;
6122
6123	shp = NFSSESSIONHASH(nd->nd_sessionid);
6124	NFSLOCKSTATE();
6125	NFSLOCKSESSION(shp);
6126	sep = nfsrv_findsession(nd->nd_sessionid);
6127	if (sep == NULL) {
6128		NFSUNLOCKSESSION(shp);
6129		NFSUNLOCKSTATE();
6130		return (NFSERR_BADSESSION);
6131	}
6132
6133	if (onefs != 0)
6134		sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6135		/* Check to see if reclaim complete has already happened. */
6136	else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6137		error = NFSERR_COMPLETEALREADY;
6138	else {
6139		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6140		nfsrv_markreclaim(sep->sess_clp);
6141	}
6142	NFSUNLOCKSESSION(shp);
6143	NFSUNLOCKSTATE();
6144	return (error);
6145}
6146
6147/*
6148 * Cache the reply in a session slot.
6149 */
6150void
6151nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
6152   struct mbuf **m)
6153{
6154	struct nfsdsession *sep;
6155	struct nfssessionhash *shp;
6156
6157	shp = NFSSESSIONHASH(sessionid);
6158	NFSLOCKSESSION(shp);
6159	sep = nfsrv_findsession(sessionid);
6160	if (sep == NULL) {
6161		NFSUNLOCKSESSION(shp);
6162		printf("nfsrv_cache_session: no session\n");
6163		m_freem(*m);
6164		return;
6165	}
6166	nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
6167	NFSUNLOCKSESSION(shp);
6168}
6169
6170/*
6171 * Search for a session that matches the sessionid.
6172 */
6173static struct nfsdsession *
6174nfsrv_findsession(uint8_t *sessionid)
6175{
6176	struct nfsdsession *sep;
6177	struct nfssessionhash *shp;
6178
6179	shp = NFSSESSIONHASH(sessionid);
6180	LIST_FOREACH(sep, &shp->list, sess_hash) {
6181		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6182			break;
6183	}
6184	return (sep);
6185}
6186
6187/*
6188 * Destroy a session.
6189 */
6190int
6191nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6192{
6193	int error, igotlock, samesess;
6194
6195	samesess = 0;
6196	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6197	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6198		samesess = 1;
6199		if ((nd->nd_flag & ND_LASTOP) == 0)
6200			return (NFSERR_BADSESSION);
6201	}
6202
6203	/* Lock out other nfsd threads */
6204	NFSLOCKV4ROOTMUTEX();
6205	nfsv4_relref(&nfsv4rootfs_lock);
6206	do {
6207		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6208		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6209	} while (igotlock == 0);
6210	NFSUNLOCKV4ROOTMUTEX();
6211
6212	error = nfsrv_freesession(NULL, sessionid);
6213	if (error == 0 && samesess != 0)
6214		nd->nd_flag &= ~ND_HASSEQUENCE;
6215
6216	NFSLOCKV4ROOTMUTEX();
6217	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6218	NFSUNLOCKV4ROOTMUTEX();
6219	return (error);
6220}
6221
6222/*
6223 * Bind a connection to a session.
6224 * For now, only certain variants are supported, since the current session
6225 * structure can only handle a single backchannel entry, which will be
6226 * applied to all connections if it is set.
6227 */
6228int
6229nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6230{
6231	struct nfssessionhash *shp;
6232	struct nfsdsession *sep;
6233	struct nfsclient *clp;
6234	SVCXPRT *savxprt;
6235	int error;
6236
6237	error = 0;
6238	shp = NFSSESSIONHASH(sessionid);
6239	NFSLOCKSTATE();
6240	NFSLOCKSESSION(shp);
6241	sep = nfsrv_findsession(sessionid);
6242	if (sep != NULL) {
6243		clp = sep->sess_clp;
6244		if (*foreaftp == NFSCDFC4_BACK ||
6245		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6246		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6247			/* Try to set up a backchannel. */
6248			if (clp->lc_req.nr_client == NULL) {
6249				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6250				    "backchannel\n");
6251				clp->lc_req.nr_client = (struct __rpc_client *)
6252				    clnt_bck_create(nd->nd_xprt->xp_socket,
6253				    sep->sess_cbprogram, NFSV4_CBVERS);
6254			}
6255			if (clp->lc_req.nr_client != NULL) {
6256				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6257				    "backchannel\n");
6258				savxprt = sep->sess_cbsess.nfsess_xprt;
6259				SVC_ACQUIRE(nd->nd_xprt);
6260				nd->nd_xprt->xp_p2 =
6261				    clp->lc_req.nr_client->cl_private;
6262				/* Disable idle timeout. */
6263				nd->nd_xprt->xp_idletimeout = 0;
6264				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6265				if (savxprt != NULL)
6266					SVC_RELEASE(savxprt);
6267				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6268				clp->lc_flags |= LCL_DONEBINDCONN;
6269				if (*foreaftp == NFSCDFS4_BACK)
6270					*foreaftp = NFSCDFS4_BACK;
6271				else
6272					*foreaftp = NFSCDFS4_BOTH;
6273			} else if (*foreaftp != NFSCDFC4_BACK) {
6274				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6275				    "up backchannel\n");
6276				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6277				clp->lc_flags |= LCL_DONEBINDCONN;
6278				*foreaftp = NFSCDFS4_FORE;
6279			} else {
6280				error = NFSERR_NOTSUPP;
6281				printf("nfsrv_bindconnsess: Can't add "
6282				    "backchannel\n");
6283			}
6284		} else {
6285			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6286			clp->lc_flags |= LCL_DONEBINDCONN;
6287			*foreaftp = NFSCDFS4_FORE;
6288		}
6289	} else
6290		error = NFSERR_BADSESSION;
6291	NFSUNLOCKSESSION(shp);
6292	NFSUNLOCKSTATE();
6293	return (error);
6294}
6295
6296/*
6297 * Free up a session structure.
6298 */
6299static int
6300nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
6301{
6302	struct nfssessionhash *shp;
6303	int i;
6304
6305	NFSLOCKSTATE();
6306	if (sep == NULL) {
6307		shp = NFSSESSIONHASH(sessionid);
6308		NFSLOCKSESSION(shp);
6309		sep = nfsrv_findsession(sessionid);
6310	} else {
6311		shp = NFSSESSIONHASH(sep->sess_sessionid);
6312		NFSLOCKSESSION(shp);
6313	}
6314	if (sep != NULL) {
6315		sep->sess_refcnt--;
6316		if (sep->sess_refcnt > 0) {
6317			NFSUNLOCKSESSION(shp);
6318			NFSUNLOCKSTATE();
6319			return (NFSERR_BACKCHANBUSY);
6320		}
6321		LIST_REMOVE(sep, sess_hash);
6322		LIST_REMOVE(sep, sess_list);
6323	}
6324	NFSUNLOCKSESSION(shp);
6325	NFSUNLOCKSTATE();
6326	if (sep == NULL)
6327		return (NFSERR_BADSESSION);
6328	for (i = 0; i < NFSV4_SLOTS; i++)
6329		if (sep->sess_slots[i].nfssl_reply != NULL)
6330			m_freem(sep->sess_slots[i].nfssl_reply);
6331	if (sep->sess_cbsess.nfsess_xprt != NULL)
6332		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6333	free(sep, M_NFSDSESSION);
6334	return (0);
6335}
6336
6337/*
6338 * Free a stateid.
6339 * RFC5661 says that it should fail when there are associated opens, locks
6340 * or delegations. Since stateids represent opens, I don't see how you can
6341 * free an open stateid (it will be free'd when closed), so this function
6342 * only works for lock stateids (freeing the lock_owner) or delegations.
6343 */
6344int
6345nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6346    NFSPROC_T *p)
6347{
6348	struct nfsclient *clp;
6349	struct nfsstate *stp;
6350	int error;
6351
6352	NFSLOCKSTATE();
6353	/*
6354	 * Look up the stateid
6355	 */
6356	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6357	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6358	if (error == 0) {
6359		/* First, check for a delegation. */
6360		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6361			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6362			    NFSX_STATEIDOTHER))
6363				break;
6364		}
6365		if (stp != NULL) {
6366			nfsrv_freedeleg(stp);
6367			NFSUNLOCKSTATE();
6368			return (error);
6369		}
6370	}
6371	/* Not a delegation, try for a lock_owner. */
6372	if (error == 0)
6373		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6374	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6375	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6376		/* Not a lock_owner stateid. */
6377		error = NFSERR_LOCKSHELD;
6378	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6379		error = NFSERR_LOCKSHELD;
6380	if (error == 0)
6381		nfsrv_freelockowner(stp, NULL, 0, p);
6382	NFSUNLOCKSTATE();
6383	return (error);
6384}
6385
6386/*
6387 * Test a stateid.
6388 */
6389int
6390nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6391    NFSPROC_T *p)
6392{
6393	struct nfsclient *clp;
6394	struct nfsstate *stp;
6395	int error;
6396
6397	NFSLOCKSTATE();
6398	/*
6399	 * Look up the stateid
6400	 */
6401	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6402	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6403	if (error == 0)
6404		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6405	if (error == 0 && stateidp->seqid != 0 &&
6406	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6407		error = NFSERR_OLDSTATEID;
6408	NFSUNLOCKSTATE();
6409	return (error);
6410}
6411
6412/*
6413 * Generate the xdr for an NFSv4.1 CBSequence Operation.
6414 */
6415static int
6416nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6417    int dont_replycache, struct nfsdsession **sepp)
6418{
6419	struct nfsdsession *sep;
6420	uint32_t *tl, slotseq = 0;
6421	int maxslot, slotpos;
6422	uint8_t sessionid[NFSX_V4SESSIONID];
6423	int error;
6424
6425	error = nfsv4_getcbsession(clp, sepp);
6426	if (error != 0)
6427		return (error);
6428	sep = *sepp;
6429	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
6430	    &slotseq, sessionid);
6431	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6432
6433	/* Build the Sequence arguments. */
6434	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6435	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6436	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6437	nd->nd_slotseq = tl;
6438	*tl++ = txdr_unsigned(slotseq);
6439	*tl++ = txdr_unsigned(slotpos);
6440	*tl++ = txdr_unsigned(maxslot);
6441	if (dont_replycache == 0)
6442		*tl++ = newnfs_true;
6443	else
6444		*tl++ = newnfs_false;
6445	*tl = 0;			/* No referring call list, for now. */
6446	nd->nd_flag |= ND_HASSEQUENCE;
6447	return (0);
6448}
6449
6450/*
6451 * Get a session for the callback.
6452 */
6453static int
6454nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6455{
6456	struct nfsdsession *sep;
6457
6458	NFSLOCKSTATE();
6459	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6460		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6461			break;
6462	}
6463	if (sep == NULL) {
6464		NFSUNLOCKSTATE();
6465		return (NFSERR_BADSESSION);
6466	}
6467	sep->sess_refcnt++;
6468	*sepp = sep;
6469	NFSUNLOCKSTATE();
6470	return (0);
6471}
6472
6473/*
6474 * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6475 * exit, since those transports will all be going away.
6476 * This is only called after all the nfsd threads are done performing RPCs,
6477 * so locking shouldn't be an issue.
6478 */
6479APPLESTATIC void
6480nfsrv_freeallbackchannel_xprts(void)
6481{
6482	struct nfsdsession *sep;
6483	struct nfsclient *clp;
6484	SVCXPRT *xprt;
6485	int i;
6486
6487	for (i = 0; i < nfsrv_clienthashsize; i++) {
6488		LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6489			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6490				xprt = sep->sess_cbsess.nfsess_xprt;
6491				sep->sess_cbsess.nfsess_xprt = NULL;
6492				if (xprt != NULL)
6493					SVC_RELEASE(xprt);
6494			}
6495		}
6496	}
6497}
6498
6499