Home | History | Annotate | Line # | Download | only in server
nfs_nfsdstate.c revision 1.1.1.1
      1 /*	$NetBSD: nfs_nfsdstate.c,v 1.1.1.1 2013/09/30 07:19:52 dholland Exp $	*/
      2 /*-
      3  * Copyright (c) 2009 Rick Macklem, University of Guelph
      4  * All rights reserved.
      5  *
      6  * Redistribution and use in source and binary forms, with or without
      7  * modification, are permitted provided that the following conditions
      8  * are met:
      9  * 1. Redistributions of source code must retain the above copyright
     10  *    notice, this list of conditions and the following disclaimer.
     11  * 2. Redistributions in binary form must reproduce the above copyright
     12  *    notice, this list of conditions and the following disclaimer in the
     13  *    documentation and/or other materials provided with the distribution.
     14  *
     15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
     16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     25  * SUCH DAMAGE.
     26  *
     27  */
     28 
     29 #include <sys/cdefs.h>
     30 /* __FBSDID("FreeBSD: head/sys/fs/nfsserver/nfs_nfsdstate.c 245909 2013-01-25 15:25:24Z jhb "); */
     31 __RCSID("$NetBSD: nfs_nfsdstate.c,v 1.1.1.1 2013/09/30 07:19:52 dholland Exp $");
     32 
     33 #ifndef APPLEKEXT
     34 #include <fs/nfs/nfsport.h>
     35 
     36 struct nfsrv_stablefirst nfsrv_stablefirst;
     37 int nfsrv_issuedelegs = 0;
     38 int nfsrv_dolocallocks = 0;
     39 struct nfsv4lock nfsv4rootfs_lock;
     40 
     41 extern int newnfs_numnfsd;
     42 extern struct nfsstats newnfsstats;
     43 extern int nfsrv_lease;
     44 extern struct timeval nfsboottime;
     45 extern u_int32_t newnfs_true, newnfs_false;
     46 NFSV4ROOTLOCKMUTEX;
     47 NFSSTATESPINLOCK;
     48 
     49 /*
     50  * Hash lists for nfs V4.
     51  * (Some would put them in the .h file, but I don't like declaring storage
     52  *  in a .h)
     53  */
     54 struct nfsclienthashhead nfsclienthash[NFSCLIENTHASHSIZE];
     55 struct nfslockhashhead nfslockhash[NFSLOCKHASHSIZE];
     56 #endif	/* !APPLEKEXT */
     57 
     58 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
     59 static time_t nfsrvboottime;
     60 static int nfsrv_writedelegifpos = 1;
     61 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
     62 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
     63 static int nfsrv_nogsscallback = 0;
     64 
     65 /* local functions */
     66 static void nfsrv_dumpaclient(struct nfsclient *clp,
     67     struct nfsd_dumpclients *dumpp);
     68 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
     69     NFSPROC_T *p);
     70 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
     71     NFSPROC_T *p);
     72 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
     73     NFSPROC_T *p);
     74 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
     75     int cansleep, NFSPROC_T *p);
     76 static void nfsrv_freenfslock(struct nfslock *lop);
     77 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
     78 static void nfsrv_freedeleg(struct nfsstate *);
     79 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
     80     u_int32_t flags, struct nfsstate **stpp);
     81 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
     82     struct nfsstate **stpp);
     83 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
     84     struct nfslockfile **new_lfpp, fhandle_t *nfhp, NFSPROC_T *p);
     85 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
     86     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
     87 static void nfsrv_insertlock(struct nfslock *new_lop,
     88     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
     89 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
     90     struct nfslock **other_lopp, struct nfslockfile *lfp);
     91 static int nfsrv_getipnumber(u_char *cp);
     92 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
     93     nfsv4stateid_t *stateidp, int specialid);
     94 static int nfsrv_checkgrace(u_int32_t flags);
     95 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
     96     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
     97     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
     98 static u_int32_t nfsrv_nextclientindex(void);
     99 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
    100 static void nfsrv_markstable(struct nfsclient *clp);
    101 static int nfsrv_checkstable(struct nfsclient *clp);
    102 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
    103     vnode *vp, NFSPROC_T *p);
    104 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
    105     NFSPROC_T *p, vnode_t vp);
    106 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
    107     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
    108 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
    109     struct nfsclient *clp);
    110 static time_t nfsrv_leaseexpiry(void);
    111 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
    112 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
    113     struct nfsstate *stp, struct nfsrvcache *op);
    114 static int nfsrv_nootherstate(struct nfsstate *stp);
    115 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
    116     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
    117 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
    118     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
    119 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
    120     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
    121     NFSPROC_T *p);
    122 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
    123     NFSPROC_T *p);
    124 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
    125     uint64_t first, uint64_t end);
    126 static void nfsrv_locklf(struct nfslockfile *lfp);
    127 static void nfsrv_unlocklf(struct nfslockfile *lfp);
    128 
    129 /*
    130  * Scan the client list for a match and either return the current one,
    131  * create a new entry or return an error.
    132  * If returning a non-error, the clp structure must either be linked into
    133  * the client list or free'd.
    134  */
    135 APPLESTATIC int
    136 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
    137     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
    138 {
    139 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
    140 	int i, error = 0;
    141 	struct nfsstate *stp, *tstp;
    142 	struct sockaddr_in *sad, *rad;
    143 	int zapit = 0, gotit, hasstate = 0, igotlock;
    144 	static u_int64_t confirm_index = 0;
    145 
    146 	/*
    147 	 * Check for state resource limit exceeded.
    148 	 */
    149 	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
    150 		error = NFSERR_RESOURCE;
    151 		goto out;
    152 	}
    153 
    154 	if (nfsrv_issuedelegs == 0 ||
    155 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
    156 		/*
    157 		 * Don't do callbacks when delegations are disabled or
    158 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
    159 		 * If establishing a callback connection is attempted
    160 		 * when a firewall is blocking the callback path, the
    161 		 * server may wait too long for the connect attempt to
    162 		 * succeed during the Open. Some clients, such as Linux,
    163 		 * may timeout and give up on the Open before the server
    164 		 * replies. Also, since AUTH_GSS callbacks are not
    165 		 * yet interoperability tested, they might cause the
    166 		 * server to crap out, if they get past the Init call to
    167 		 * the client.
    168 		 */
    169 		new_clp->lc_program = 0;
    170 
    171 	/* Lock out other nfsd threads */
    172 	NFSLOCKV4ROOTMUTEX();
    173 	nfsv4_relref(&nfsv4rootfs_lock);
    174 	do {
    175 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
    176 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
    177 	} while (!igotlock);
    178 	NFSUNLOCKV4ROOTMUTEX();
    179 
    180 	/*
    181 	 * Search for a match in the client list.
    182 	 */
    183 	gotit = i = 0;
    184 	while (i < NFSCLIENTHASHSIZE && !gotit) {
    185 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
    186 		if (new_clp->lc_idlen == clp->lc_idlen &&
    187 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
    188 			gotit = 1;
    189 			break;
    190 		}
    191 	    }
    192 	    i++;
    193 	}
    194 	if (!gotit ||
    195 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
    196 		/*
    197 		 * Get rid of the old one.
    198 		 */
    199 		if (i != NFSCLIENTHASHSIZE) {
    200 			LIST_REMOVE(clp, lc_hash);
    201 			nfsrv_cleanclient(clp, p);
    202 			nfsrv_freedeleglist(&clp->lc_deleg);
    203 			nfsrv_freedeleglist(&clp->lc_olddeleg);
    204 			zapit = 1;
    205 		}
    206 		/*
    207 		 * Add it after assigning a client id to it.
    208 		 */
    209 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
    210 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
    211 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
    212 		    (u_int32_t)nfsrvboottime;
    213 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
    214 		    nfsrv_nextclientindex();
    215 		new_clp->lc_stateindex = 0;
    216 		new_clp->lc_statemaxindex = 0;
    217 		new_clp->lc_cbref = 0;
    218 		new_clp->lc_expiry = nfsrv_leaseexpiry();
    219 		LIST_INIT(&new_clp->lc_open);
    220 		LIST_INIT(&new_clp->lc_deleg);
    221 		LIST_INIT(&new_clp->lc_olddeleg);
    222 		for (i = 0; i < NFSSTATEHASHSIZE; i++)
    223 			LIST_INIT(&new_clp->lc_stateid[i]);
    224 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
    225 		    lc_hash);
    226 		newnfsstats.srvclients++;
    227 		nfsrv_openpluslock++;
    228 		nfsrv_clients++;
    229 		NFSLOCKV4ROOTMUTEX();
    230 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
    231 		NFSUNLOCKV4ROOTMUTEX();
    232 		if (zapit)
    233 			nfsrv_zapclient(clp, p);
    234 		*new_clpp = NULL;
    235 		goto out;
    236 	}
    237 
    238 	/*
    239 	 * Now, handle the cases where the id is already issued.
    240 	 */
    241 	if (nfsrv_notsamecredname(nd, clp)) {
    242 	    /*
    243 	     * Check to see if there is expired state that should go away.
    244 	     */
    245 	    if (clp->lc_expiry < NFSD_MONOSEC &&
    246 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
    247 		nfsrv_cleanclient(clp, p);
    248 		nfsrv_freedeleglist(&clp->lc_deleg);
    249 	    }
    250 
    251 	    /*
    252 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
    253 	     * RFC3530 Sec. 8.1.2 last para.
    254 	     */
    255 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
    256 		hasstate = 1;
    257 	    } else if (LIST_EMPTY(&clp->lc_open)) {
    258 		hasstate = 0;
    259 	    } else {
    260 		hasstate = 0;
    261 		/* Look for an Open on the OpenOwner */
    262 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
    263 		    if (!LIST_EMPTY(&stp->ls_open)) {
    264 			hasstate = 1;
    265 			break;
    266 		    }
    267 		}
    268 	    }
    269 	    if (hasstate) {
    270 		/*
    271 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
    272 		 * filling out the correct ipaddr and portnum.
    273 		 */
    274 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
    275 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
    276 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
    277 		sad->sin_port = rad->sin_port;
    278 		NFSLOCKV4ROOTMUTEX();
    279 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
    280 		NFSUNLOCKV4ROOTMUTEX();
    281 		error = NFSERR_CLIDINUSE;
    282 		goto out;
    283 	    }
    284 	}
    285 
    286 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
    287 		/*
    288 		 * If the verifier has changed, the client has rebooted
    289 		 * and a new client id is issued. The old state info
    290 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
    291 		 */
    292 		LIST_REMOVE(clp, lc_hash);
    293 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
    294 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
    295 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
    296 		    nfsrvboottime;
    297 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
    298 		    nfsrv_nextclientindex();
    299 		new_clp->lc_stateindex = 0;
    300 		new_clp->lc_statemaxindex = 0;
    301 		new_clp->lc_cbref = 0;
    302 		new_clp->lc_expiry = nfsrv_leaseexpiry();
    303 
    304 		/*
    305 		 * Save the state until confirmed.
    306 		 */
    307 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
    308 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
    309 			tstp->ls_clp = new_clp;
    310 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
    311 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
    312 			tstp->ls_clp = new_clp;
    313 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
    314 		    ls_list);
    315 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
    316 			tstp->ls_clp = new_clp;
    317 		for (i = 0; i < NFSSTATEHASHSIZE; i++) {
    318 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
    319 			    &clp->lc_stateid[i], ls_hash);
    320 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
    321 				tstp->ls_clp = new_clp;
    322 		}
    323 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
    324 		    lc_hash);
    325 		newnfsstats.srvclients++;
    326 		nfsrv_openpluslock++;
    327 		nfsrv_clients++;
    328 		NFSLOCKV4ROOTMUTEX();
    329 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
    330 		NFSUNLOCKV4ROOTMUTEX();
    331 
    332 		/*
    333 		 * Must wait until any outstanding callback on the old clp
    334 		 * completes.
    335 		 */
    336 		NFSLOCKSTATE();
    337 		while (clp->lc_cbref) {
    338 			clp->lc_flags |= LCL_WAKEUPWANTED;
    339 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
    340 			    "nfsd clp", 10 * hz);
    341 		}
    342 		NFSUNLOCKSTATE();
    343 		nfsrv_zapclient(clp, p);
    344 		*new_clpp = NULL;
    345 		goto out;
    346 	}
    347 	/*
    348 	 * id and verifier match, so update the net address info
    349 	 * and get rid of any existing callback authentication
    350 	 * handle, so a new one will be acquired.
    351 	 */
    352 	LIST_REMOVE(clp, lc_hash);
    353 	new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
    354 	new_clp->lc_expiry = nfsrv_leaseexpiry();
    355 	confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
    356 	clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
    357 	    clp->lc_clientid.lval[0];
    358 	clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
    359 	    clp->lc_clientid.lval[1];
    360 	new_clp->lc_delegtime = clp->lc_delegtime;
    361 	new_clp->lc_stateindex = clp->lc_stateindex;
    362 	new_clp->lc_statemaxindex = clp->lc_statemaxindex;
    363 	new_clp->lc_cbref = 0;
    364 	LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
    365 	LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
    366 		tstp->ls_clp = new_clp;
    367 	LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
    368 	LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
    369 		tstp->ls_clp = new_clp;
    370 	LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
    371 	LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
    372 		tstp->ls_clp = new_clp;
    373 	for (i = 0; i < NFSSTATEHASHSIZE; i++) {
    374 		LIST_NEWHEAD(&new_clp->lc_stateid[i], &clp->lc_stateid[i],
    375 		    ls_hash);
    376 		LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
    377 			tstp->ls_clp = new_clp;
    378 	}
    379 	LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
    380 	    lc_hash);
    381 	newnfsstats.srvclients++;
    382 	nfsrv_openpluslock++;
    383 	nfsrv_clients++;
    384 	NFSLOCKV4ROOTMUTEX();
    385 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
    386 	NFSUNLOCKV4ROOTMUTEX();
    387 
    388 	/*
    389 	 * Must wait until any outstanding callback on the old clp
    390 	 * completes.
    391 	 */
    392 	NFSLOCKSTATE();
    393 	while (clp->lc_cbref) {
    394 		clp->lc_flags |= LCL_WAKEUPWANTED;
    395 		(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1, "nfsd clp",
    396 		    10 * hz);
    397 	}
    398 	NFSUNLOCKSTATE();
    399 	nfsrv_zapclient(clp, p);
    400 	*new_clpp = NULL;
    401 
    402 out:
    403 	NFSEXITCODE2(error, nd);
    404 	return (error);
    405 }
    406 
    407 /*
    408  * Check to see if the client id exists and optionally confirm it.
    409  */
    410 APPLESTATIC int
    411 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
    412     nfsquad_t confirm, struct nfsrv_descript *nd, NFSPROC_T *p)
    413 {
    414 	struct nfsclient *clp;
    415 	struct nfsstate *stp;
    416 	int i;
    417 	struct nfsclienthashhead *hp;
    418 	int error = 0, igotlock, doneok;
    419 
    420 	if (clpp)
    421 		*clpp = NULL;
    422 	if (nfsrvboottime != clientid.lval[0]) {
    423 		error = NFSERR_STALECLIENTID;
    424 		goto out;
    425 	}
    426 
    427 	/*
    428 	 * If called with opflags == CLOPS_RENEW, the State Lock is
    429 	 * already held. Otherwise, we need to get either that or,
    430 	 * for the case of Confirm, lock out the nfsd threads.
    431 	 */
    432 	if (opflags & CLOPS_CONFIRM) {
    433 		NFSLOCKV4ROOTMUTEX();
    434 		nfsv4_relref(&nfsv4rootfs_lock);
    435 		do {
    436 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
    437 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
    438 		} while (!igotlock);
    439 		NFSUNLOCKV4ROOTMUTEX();
    440 	} else if (opflags != CLOPS_RENEW) {
    441 		NFSLOCKSTATE();
    442 	}
    443 
    444 	hp = NFSCLIENTHASH(clientid);
    445 	LIST_FOREACH(clp, hp, lc_hash) {
    446 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
    447 			break;
    448 	}
    449 	if (clp == LIST_END(hp)) {
    450 		if (opflags & CLOPS_CONFIRM)
    451 			error = NFSERR_STALECLIENTID;
    452 		else
    453 			error = NFSERR_EXPIRED;
    454 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
    455 		/*
    456 		 * If marked admin revoked, just return the error.
    457 		 */
    458 		error = NFSERR_ADMINREVOKED;
    459 	}
    460 	if (error) {
    461 		if (opflags & CLOPS_CONFIRM) {
    462 			NFSLOCKV4ROOTMUTEX();
    463 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
    464 			NFSUNLOCKV4ROOTMUTEX();
    465 		} else if (opflags != CLOPS_RENEW) {
    466 			NFSUNLOCKSTATE();
    467 		}
    468 		goto out;
    469 	}
    470 
    471 	/*
    472 	 * Perform any operations specified by the opflags.
    473 	 */
    474 	if (opflags & CLOPS_CONFIRM) {
    475 		if (clp->lc_confirm.qval != confirm.qval)
    476 			error = NFSERR_STALECLIENTID;
    477 		else if (nfsrv_notsamecredname(nd, clp))
    478 			error = NFSERR_CLIDINUSE;
    479 
    480 		if (!error) {
    481 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
    482 			LCL_NEEDSCONFIRM) {
    483 			/*
    484 			 * Hang onto the delegations (as old delegations)
    485 			 * for an Open with CLAIM_DELEGATE_PREV unless in
    486 			 * grace, but get rid of the rest of the state.
    487 			 */
    488 			nfsrv_cleanclient(clp, p);
    489 			nfsrv_freedeleglist(&clp->lc_olddeleg);
    490 			if (nfsrv_checkgrace(0)) {
    491 			    /* In grace, so just delete delegations */
    492 			    nfsrv_freedeleglist(&clp->lc_deleg);
    493 			} else {
    494 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
    495 				stp->ls_flags |= NFSLCK_OLDDELEG;
    496 			    clp->lc_delegtime = NFSD_MONOSEC +
    497 				nfsrv_lease + NFSRV_LEASEDELTA;
    498 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
    499 				ls_list);
    500 			}
    501 		    }
    502 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
    503 		    if (clp->lc_program)
    504 			clp->lc_flags |= LCL_NEEDSCBNULL;
    505 		}
    506 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
    507 		error = NFSERR_EXPIRED;
    508 	}
    509 
    510 	/*
    511 	 * If called by the Renew Op, we must check the principal.
    512 	 */
    513 	if (!error && (opflags & CLOPS_RENEWOP)) {
    514 	    if (nfsrv_notsamecredname(nd, clp)) {
    515 		doneok = 0;
    516 		for (i = 0; i < NFSSTATEHASHSIZE && doneok == 0; i++) {
    517 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
    518 			if ((stp->ls_flags & NFSLCK_OPEN) &&
    519 			    stp->ls_uid == nd->nd_cred->cr_uid) {
    520 				doneok = 1;
    521 				break;
    522 			}
    523 		    }
    524 		}
    525 		if (!doneok)
    526 			error = NFSERR_ACCES;
    527 	    }
    528 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
    529 		error = NFSERR_CBPATHDOWN;
    530 	}
    531 	if ((!error || error == NFSERR_CBPATHDOWN) &&
    532 	     (opflags & CLOPS_RENEW)) {
    533 		clp->lc_expiry = nfsrv_leaseexpiry();
    534 	}
    535 	if (opflags & CLOPS_CONFIRM) {
    536 		NFSLOCKV4ROOTMUTEX();
    537 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
    538 		NFSUNLOCKV4ROOTMUTEX();
    539 	} else if (opflags != CLOPS_RENEW) {
    540 		NFSUNLOCKSTATE();
    541 	}
    542 	if (clpp)
    543 		*clpp = clp;
    544 
    545 out:
    546 	NFSEXITCODE2(error, nd);
    547 	return (error);
    548 }
    549 
    550 /*
    551  * Called from the new nfssvc syscall to admin revoke a clientid.
    552  * Returns 0 for success, error otherwise.
    553  */
    554 APPLESTATIC int
    555 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
    556 {
    557 	struct nfsclient *clp = NULL;
    558 	int i, error = 0;
    559 	int gotit, igotlock;
    560 
    561 	/*
    562 	 * First, lock out the nfsd so that state won't change while the
    563 	 * revocation record is being written to the stable storage restart
    564 	 * file.
    565 	 */
    566 	NFSLOCKV4ROOTMUTEX();
    567 	do {
    568 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
    569 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
    570 	} while (!igotlock);
    571 	NFSUNLOCKV4ROOTMUTEX();
    572 
    573 	/*
    574 	 * Search for a match in the client list.
    575 	 */
    576 	gotit = i = 0;
    577 	while (i < NFSCLIENTHASHSIZE && !gotit) {
    578 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
    579 		if (revokep->nclid_idlen == clp->lc_idlen &&
    580 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
    581 			gotit = 1;
    582 			break;
    583 		}
    584 	    }
    585 	    i++;
    586 	}
    587 	if (!gotit) {
    588 		NFSLOCKV4ROOTMUTEX();
    589 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
    590 		NFSUNLOCKV4ROOTMUTEX();
    591 		error = EPERM;
    592 		goto out;
    593 	}
    594 
    595 	/*
    596 	 * Now, write out the revocation record
    597 	 */
    598 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
    599 	nfsrv_backupstable();
    600 
    601 	/*
    602 	 * and clear out the state, marking the clientid revoked.
    603 	 */
    604 	clp->lc_flags &= ~LCL_CALLBACKSON;
    605 	clp->lc_flags |= LCL_ADMINREVOKED;
    606 	nfsrv_cleanclient(clp, p);
    607 	nfsrv_freedeleglist(&clp->lc_deleg);
    608 	nfsrv_freedeleglist(&clp->lc_olddeleg);
    609 	NFSLOCKV4ROOTMUTEX();
    610 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
    611 	NFSUNLOCKV4ROOTMUTEX();
    612 
    613 out:
    614 	NFSEXITCODE(error);
    615 	return (error);
    616 }
    617 
    618 /*
    619  * Dump out stats for all clients. Called from nfssvc(2), that is used
    620  * newnfsstats.
    621  */
    622 APPLESTATIC void
    623 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
    624 {
    625 	struct nfsclient *clp;
    626 	int i = 0, cnt = 0;
    627 
    628 	/*
    629 	 * First, get a reference on the nfsv4rootfs_lock so that an
    630 	 * exclusive lock cannot be acquired while dumping the clients.
    631 	 */
    632 	NFSLOCKV4ROOTMUTEX();
    633 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
    634 	NFSUNLOCKV4ROOTMUTEX();
    635 	NFSLOCKSTATE();
    636 	/*
    637 	 * Rattle through the client lists until done.
    638 	 */
    639 	while (i < NFSCLIENTHASHSIZE && cnt < maxcnt) {
    640 	    clp = LIST_FIRST(&nfsclienthash[i]);
    641 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
    642 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
    643 		cnt++;
    644 		clp = LIST_NEXT(clp, lc_hash);
    645 	    }
    646 	    i++;
    647 	}
    648 	if (cnt < maxcnt)
    649 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
    650 	NFSUNLOCKSTATE();
    651 	NFSLOCKV4ROOTMUTEX();
    652 	nfsv4_relref(&nfsv4rootfs_lock);
    653 	NFSUNLOCKV4ROOTMUTEX();
    654 }
    655 
    656 /*
    657  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
    658  */
    659 static void
    660 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
    661 {
    662 	struct nfsstate *stp, *openstp, *lckownstp;
    663 	struct nfslock *lop;
    664 	struct sockaddr *sad;
    665 	struct sockaddr_in *rad;
    666 	struct sockaddr_in6 *rad6;
    667 
    668 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
    669 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
    670 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
    671 	dumpp->ndcl_flags = clp->lc_flags;
    672 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
    673 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
    674 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
    675 	dumpp->ndcl_addrfam = sad->sa_family;
    676 	if (sad->sa_family == AF_INET) {
    677 		rad = (struct sockaddr_in *)sad;
    678 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
    679 	} else {
    680 		rad6 = (struct sockaddr_in6 *)sad;
    681 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
    682 	}
    683 
    684 	/*
    685 	 * Now, scan the state lists and total up the opens and locks.
    686 	 */
    687 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
    688 	    dumpp->ndcl_nopenowners++;
    689 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
    690 		dumpp->ndcl_nopens++;
    691 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
    692 		    dumpp->ndcl_nlockowners++;
    693 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
    694 			dumpp->ndcl_nlocks++;
    695 		    }
    696 		}
    697 	    }
    698 	}
    699 
    700 	/*
    701 	 * and the delegation lists.
    702 	 */
    703 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
    704 	    dumpp->ndcl_ndelegs++;
    705 	}
    706 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
    707 	    dumpp->ndcl_nolddelegs++;
    708 	}
    709 }
    710 
    711 /*
    712  * Dump out lock stats for a file.
    713  */
    714 APPLESTATIC void
    715 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
    716     NFSPROC_T *p)
    717 {
    718 	struct nfsstate *stp;
    719 	struct nfslock *lop;
    720 	int cnt = 0;
    721 	struct nfslockfile *lfp;
    722 	struct sockaddr *sad;
    723 	struct sockaddr_in *rad;
    724 	struct sockaddr_in6 *rad6;
    725 	int ret;
    726 	fhandle_t nfh;
    727 
    728 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
    729 	/*
    730 	 * First, get a reference on the nfsv4rootfs_lock so that an
    731 	 * exclusive lock on it cannot be acquired while dumping the locks.
    732 	 */
    733 	NFSLOCKV4ROOTMUTEX();
    734 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
    735 	NFSUNLOCKV4ROOTMUTEX();
    736 	NFSLOCKSTATE();
    737 	if (!ret)
    738 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
    739 	if (ret) {
    740 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
    741 		NFSUNLOCKSTATE();
    742 		NFSLOCKV4ROOTMUTEX();
    743 		nfsv4_relref(&nfsv4rootfs_lock);
    744 		NFSUNLOCKV4ROOTMUTEX();
    745 		return;
    746 	}
    747 
    748 	/*
    749 	 * For each open share on file, dump it out.
    750 	 */
    751 	stp = LIST_FIRST(&lfp->lf_open);
    752 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
    753 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
    754 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
    755 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
    756 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
    757 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
    758 		ldumpp[cnt].ndlck_owner.nclid_idlen =
    759 		    stp->ls_openowner->ls_ownerlen;
    760 		NFSBCOPY(stp->ls_openowner->ls_owner,
    761 		    ldumpp[cnt].ndlck_owner.nclid_id,
    762 		    stp->ls_openowner->ls_ownerlen);
    763 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
    764 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
    765 		    stp->ls_clp->lc_idlen);
    766 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
    767 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
    768 		if (sad->sa_family == AF_INET) {
    769 			rad = (struct sockaddr_in *)sad;
    770 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
    771 		} else {
    772 			rad6 = (struct sockaddr_in6 *)sad;
    773 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
    774 		}
    775 		stp = LIST_NEXT(stp, ls_file);
    776 		cnt++;
    777 	}
    778 
    779 	/*
    780 	 * and all locks.
    781 	 */
    782 	lop = LIST_FIRST(&lfp->lf_lock);
    783 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
    784 		stp = lop->lo_stp;
    785 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
    786 		ldumpp[cnt].ndlck_first = lop->lo_first;
    787 		ldumpp[cnt].ndlck_end = lop->lo_end;
    788 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
    789 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
    790 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
    791 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
    792 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
    793 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
    794 		    stp->ls_ownerlen);
    795 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
    796 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
    797 		    stp->ls_clp->lc_idlen);
    798 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
    799 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
    800 		if (sad->sa_family == AF_INET) {
    801 			rad = (struct sockaddr_in *)sad;
    802 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
    803 		} else {
    804 			rad6 = (struct sockaddr_in6 *)sad;
    805 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
    806 		}
    807 		lop = LIST_NEXT(lop, lo_lckfile);
    808 		cnt++;
    809 	}
    810 
    811 	/*
    812 	 * and the delegations.
    813 	 */
    814 	stp = LIST_FIRST(&lfp->lf_deleg);
    815 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
    816 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
    817 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
    818 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
    819 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
    820 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
    821 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
    822 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
    823 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
    824 		    stp->ls_clp->lc_idlen);
    825 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
    826 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
    827 		if (sad->sa_family == AF_INET) {
    828 			rad = (struct sockaddr_in *)sad;
    829 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
    830 		} else {
    831 			rad6 = (struct sockaddr_in6 *)sad;
    832 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
    833 		}
    834 		stp = LIST_NEXT(stp, ls_file);
    835 		cnt++;
    836 	}
    837 
    838 	/*
    839 	 * If list isn't full, mark end of list by setting the client name
    840 	 * to zero length.
    841 	 */
    842 	if (cnt < maxcnt)
    843 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
    844 	NFSUNLOCKSTATE();
    845 	NFSLOCKV4ROOTMUTEX();
    846 	nfsv4_relref(&nfsv4rootfs_lock);
    847 	NFSUNLOCKV4ROOTMUTEX();
    848 }
    849 
    850 /*
    851  * Server timer routine. It can scan any linked list, so long
    852  * as it holds the spin/mutex lock and there is no exclusive lock on
    853  * nfsv4rootfs_lock.
    854  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
    855  *  to do this from a callout, since the spin locks work. For
    856  *  Darwin, I'm not sure what will work correctly yet.)
    857  * Should be called once per second.
    858  */
    859 APPLESTATIC void
    860 nfsrv_servertimer(void)
    861 {
    862 	struct nfsclient *clp, *nclp;
    863 	struct nfsstate *stp, *nstp;
    864 	int got_ref, i;
    865 
    866 	/*
    867 	 * Make sure nfsboottime is set. This is used by V3 as well
    868 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
    869 	 * only used by the V4 server for leases.
    870 	 */
    871 	if (nfsboottime.tv_sec == 0)
    872 		NFSSETBOOTTIME(nfsboottime);
    873 
    874 	/*
    875 	 * If server hasn't started yet, just return.
    876 	 */
    877 	NFSLOCKSTATE();
    878 	if (nfsrv_stablefirst.nsf_eograce == 0) {
    879 		NFSUNLOCKSTATE();
    880 		return;
    881 	}
    882 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
    883 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
    884 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
    885 			nfsrv_stablefirst.nsf_flags |=
    886 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
    887 		NFSUNLOCKSTATE();
    888 		return;
    889 	}
    890 
    891 	/*
    892 	 * Try and get a reference count on the nfsv4rootfs_lock so that
    893 	 * no nfsd thread can acquire an exclusive lock on it before this
    894 	 * call is done. If it is already exclusively locked, just return.
    895 	 */
    896 	NFSLOCKV4ROOTMUTEX();
    897 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
    898 	NFSUNLOCKV4ROOTMUTEX();
    899 	if (got_ref == 0) {
    900 		NFSUNLOCKSTATE();
    901 		return;
    902 	}
    903 
    904 	/*
    905 	 * For each client...
    906 	 */
    907 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
    908 	    clp = LIST_FIRST(&nfsclienthash[i]);
    909 	    while (clp != LIST_END(&nfsclienthash[i])) {
    910 		nclp = LIST_NEXT(clp, lc_hash);
    911 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
    912 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
    913 			 && ((LIST_EMPTY(&clp->lc_deleg)
    914 			      && LIST_EMPTY(&clp->lc_open)) ||
    915 			     nfsrv_clients > nfsrv_clienthighwater)) ||
    916 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
    917 			(clp->lc_expiry < NFSD_MONOSEC &&
    918 			 (nfsrv_openpluslock * 10 / 9) > NFSRV_V4STATELIMIT)) {
    919 			/*
    920 			 * Lease has expired several nfsrv_lease times ago:
    921 			 * PLUS
    922 			 *    - no state is associated with it
    923 			 *    OR
    924 			 *    - above high water mark for number of clients
    925 			 *      (nfsrv_clienthighwater should be large enough
    926 			 *       that this only occurs when clients fail to
    927 			 *       use the same nfs_client_id4.id. Maybe somewhat
    928 			 *       higher that the maximum number of clients that
    929 			 *       will mount this server?)
    930 			 * OR
    931 			 * Lease has expired a very long time ago
    932 			 * OR
    933 			 * Lease has expired PLUS the number of opens + locks
    934 			 * has exceeded 90% of capacity
    935 			 *
    936 			 * --> Mark for expiry. The actual expiry will be done
    937 			 *     by an nfsd sometime soon.
    938 			 */
    939 			clp->lc_flags |= LCL_EXPIREIT;
    940 			nfsrv_stablefirst.nsf_flags |=
    941 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
    942 		    } else {
    943 			/*
    944 			 * If there are no opens, increment no open tick cnt
    945 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
    946 			 * otherwise, if there is an open, reset no open time
    947 			 * Hopefully, this will avoid excessive re-creation
    948 			 * of open owners and subsequent open confirms.
    949 			 */
    950 			stp = LIST_FIRST(&clp->lc_open);
    951 			while (stp != LIST_END(&clp->lc_open)) {
    952 				nstp = LIST_NEXT(stp, ls_list);
    953 				if (LIST_EMPTY(&stp->ls_open)) {
    954 					stp->ls_noopens++;
    955 					if (stp->ls_noopens > NFSNOOPEN ||
    956 					    (nfsrv_openpluslock * 2) >
    957 					    NFSRV_V4STATELIMIT)
    958 						nfsrv_stablefirst.nsf_flags |=
    959 							NFSNSF_NOOPENS;
    960 				} else {
    961 					stp->ls_noopens = 0;
    962 				}
    963 				stp = nstp;
    964 			}
    965 		    }
    966 		}
    967 		clp = nclp;
    968 	    }
    969 	}
    970 	NFSUNLOCKSTATE();
    971 	NFSLOCKV4ROOTMUTEX();
    972 	nfsv4_relref(&nfsv4rootfs_lock);
    973 	NFSUNLOCKV4ROOTMUTEX();
    974 }
    975 
    976 /*
    977  * The following set of functions free up the various data structures.
    978  */
    979 /*
    980  * Clear out all open/lock state related to this nfsclient.
    981  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
    982  * there are no other active nfsd threads.
    983  */
    984 APPLESTATIC void
    985 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
    986 {
    987 	struct nfsstate *stp, *nstp;
    988 
    989 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
    990 		nfsrv_freeopenowner(stp, 1, p);
    991 }
    992 
    993 /*
    994  * Free a client that has been cleaned. It should also already have been
    995  * removed from the lists.
    996  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
    997  *  softclock interrupts are enabled.)
    998  */
    999 APPLESTATIC void
   1000 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
   1001 {
   1002 
   1003 #ifdef notyet
   1004 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
   1005 	     (LCL_GSS | LCL_CALLBACKSON) &&
   1006 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
   1007 	    clp->lc_handlelen > 0) {
   1008 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
   1009 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
   1010 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
   1011 			NULL, 0, NULL, NULL, NULL, p);
   1012 	}
   1013 #endif
   1014 	newnfs_disconnect(&clp->lc_req);
   1015 	NFSSOCKADDRFREE(clp->lc_req.nr_nam);
   1016 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
   1017 	free((caddr_t)clp, M_NFSDCLIENT);
   1018 	NFSLOCKSTATE();
   1019 	newnfsstats.srvclients--;
   1020 	nfsrv_openpluslock--;
   1021 	nfsrv_clients--;
   1022 	NFSUNLOCKSTATE();
   1023 }
   1024 
   1025 /*
   1026  * Free a list of delegation state structures.
   1027  * (This function will also free all nfslockfile structures that no
   1028  *  longer have associated state.)
   1029  */
   1030 APPLESTATIC void
   1031 nfsrv_freedeleglist(struct nfsstatehead *sthp)
   1032 {
   1033 	struct nfsstate *stp, *nstp;
   1034 
   1035 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
   1036 		nfsrv_freedeleg(stp);
   1037 	}
   1038 	LIST_INIT(sthp);
   1039 }
   1040 
   1041 /*
   1042  * Free up a delegation.
   1043  */
   1044 static void
   1045 nfsrv_freedeleg(struct nfsstate *stp)
   1046 {
   1047 	struct nfslockfile *lfp;
   1048 
   1049 	LIST_REMOVE(stp, ls_hash);
   1050 	LIST_REMOVE(stp, ls_list);
   1051 	LIST_REMOVE(stp, ls_file);
   1052 	lfp = stp->ls_lfp;
   1053 	if (LIST_EMPTY(&lfp->lf_open) &&
   1054 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
   1055 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
   1056 	    lfp->lf_usecount == 0 &&
   1057 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
   1058 		nfsrv_freenfslockfile(lfp);
   1059 	FREE((caddr_t)stp, M_NFSDSTATE);
   1060 	newnfsstats.srvdelegates--;
   1061 	nfsrv_openpluslock--;
   1062 	nfsrv_delegatecnt--;
   1063 }
   1064 
   1065 /*
   1066  * This function frees an open owner and all associated opens.
   1067  */
   1068 static void
   1069 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
   1070 {
   1071 	struct nfsstate *nstp, *tstp;
   1072 
   1073 	LIST_REMOVE(stp, ls_list);
   1074 	/*
   1075 	 * Now, free all associated opens.
   1076 	 */
   1077 	nstp = LIST_FIRST(&stp->ls_open);
   1078 	while (nstp != LIST_END(&stp->ls_open)) {
   1079 		tstp = nstp;
   1080 		nstp = LIST_NEXT(nstp, ls_list);
   1081 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
   1082 	}
   1083 	if (stp->ls_op)
   1084 		nfsrvd_derefcache(stp->ls_op);
   1085 	FREE((caddr_t)stp, M_NFSDSTATE);
   1086 	newnfsstats.srvopenowners--;
   1087 	nfsrv_openpluslock--;
   1088 }
   1089 
   1090 /*
   1091  * This function frees an open (nfsstate open structure) with all associated
   1092  * lock_owners and locks. It also frees the nfslockfile structure iff there
   1093  * are no other opens on the file.
   1094  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
   1095  */
   1096 static int
   1097 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
   1098 {
   1099 	struct nfsstate *nstp, *tstp;
   1100 	struct nfslockfile *lfp;
   1101 	int ret;
   1102 
   1103 	LIST_REMOVE(stp, ls_hash);
   1104 	LIST_REMOVE(stp, ls_list);
   1105 	LIST_REMOVE(stp, ls_file);
   1106 
   1107 	lfp = stp->ls_lfp;
   1108 	/*
   1109 	 * Now, free all lockowners associated with this open.
   1110 	 */
   1111 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
   1112 		nfsrv_freelockowner(tstp, vp, cansleep, p);
   1113 
   1114 	/*
   1115 	 * The nfslockfile is freed here if there are no locks
   1116 	 * associated with the open.
   1117 	 * If there are locks associated with the open, the
   1118 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
   1119 	 * Acquire the state mutex to avoid races with calls to
   1120 	 * nfsrv_getlockfile().
   1121 	 */
   1122 	if (cansleep != 0)
   1123 		NFSLOCKSTATE();
   1124 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
   1125 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
   1126 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
   1127 	    lfp->lf_usecount == 0 &&
   1128 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
   1129 		nfsrv_freenfslockfile(lfp);
   1130 		ret = 1;
   1131 	} else
   1132 		ret = 0;
   1133 	if (cansleep != 0)
   1134 		NFSUNLOCKSTATE();
   1135 	FREE((caddr_t)stp, M_NFSDSTATE);
   1136 	newnfsstats.srvopens--;
   1137 	nfsrv_openpluslock--;
   1138 	return (ret);
   1139 }
   1140 
   1141 /*
   1142  * Frees a lockowner and all associated locks.
   1143  */
   1144 static void
   1145 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
   1146     NFSPROC_T *p)
   1147 {
   1148 
   1149 	LIST_REMOVE(stp, ls_hash);
   1150 	LIST_REMOVE(stp, ls_list);
   1151 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
   1152 	if (stp->ls_op)
   1153 		nfsrvd_derefcache(stp->ls_op);
   1154 	FREE((caddr_t)stp, M_NFSDSTATE);
   1155 	newnfsstats.srvlockowners--;
   1156 	nfsrv_openpluslock--;
   1157 }
   1158 
   1159 /*
   1160  * Free all the nfs locks on a lockowner.
   1161  */
   1162 static void
   1163 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
   1164     NFSPROC_T *p)
   1165 {
   1166 	struct nfslock *lop, *nlop;
   1167 	struct nfsrollback *rlp, *nrlp;
   1168 	struct nfslockfile *lfp = NULL;
   1169 	int gottvp = 0;
   1170 	vnode_t tvp = NULL;
   1171 	uint64_t first, end;
   1172 
   1173 	lop = LIST_FIRST(&stp->ls_lock);
   1174 	while (lop != LIST_END(&stp->ls_lock)) {
   1175 		nlop = LIST_NEXT(lop, lo_lckowner);
   1176 		/*
   1177 		 * Since all locks should be for the same file, lfp should
   1178 		 * not change.
   1179 		 */
   1180 		if (lfp == NULL)
   1181 			lfp = lop->lo_lfp;
   1182 		else if (lfp != lop->lo_lfp)
   1183 			panic("allnfslocks");
   1184 		/*
   1185 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
   1186 		 * from the file handle. This only occurs when called from
   1187 		 * nfsrv_cleanclient().
   1188 		 */
   1189 		if (gottvp == 0) {
   1190 			if (nfsrv_dolocallocks == 0)
   1191 				tvp = NULL;
   1192 			else if (vp == NULL && cansleep != 0)
   1193 				tvp = nfsvno_getvp(&lfp->lf_fh);
   1194 			else
   1195 				tvp = vp;
   1196 			gottvp = 1;
   1197 		}
   1198 
   1199 		if (tvp != NULL) {
   1200 			if (cansleep == 0)
   1201 				panic("allnfs2");
   1202 			first = lop->lo_first;
   1203 			end = lop->lo_end;
   1204 			nfsrv_freenfslock(lop);
   1205 			nfsrv_localunlock(tvp, lfp, first, end, p);
   1206 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
   1207 			    nrlp)
   1208 				free(rlp, M_NFSDROLLBACK);
   1209 			LIST_INIT(&lfp->lf_rollback);
   1210 		} else
   1211 			nfsrv_freenfslock(lop);
   1212 		lop = nlop;
   1213 	}
   1214 	if (vp == NULL && tvp != NULL)
   1215 		vput(tvp);
   1216 }
   1217 
   1218 /*
   1219  * Free an nfslock structure.
   1220  */
   1221 static void
   1222 nfsrv_freenfslock(struct nfslock *lop)
   1223 {
   1224 
   1225 	if (lop->lo_lckfile.le_prev != NULL) {
   1226 		LIST_REMOVE(lop, lo_lckfile);
   1227 		newnfsstats.srvlocks--;
   1228 		nfsrv_openpluslock--;
   1229 	}
   1230 	LIST_REMOVE(lop, lo_lckowner);
   1231 	FREE((caddr_t)lop, M_NFSDLOCK);
   1232 }
   1233 
   1234 /*
   1235  * This function frees an nfslockfile structure.
   1236  */
   1237 static void
   1238 nfsrv_freenfslockfile(struct nfslockfile *lfp)
   1239 {
   1240 
   1241 	LIST_REMOVE(lfp, lf_hash);
   1242 	FREE((caddr_t)lfp, M_NFSDLOCKFILE);
   1243 }
   1244 
   1245 /*
   1246  * This function looks up an nfsstate structure via stateid.
   1247  */
   1248 static int
   1249 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
   1250     struct nfsstate **stpp)
   1251 {
   1252 	struct nfsstate *stp;
   1253 	struct nfsstatehead *hp;
   1254 	int error = 0;
   1255 
   1256 	*stpp = NULL;
   1257 	hp = NFSSTATEHASH(clp, *stateidp);
   1258 	LIST_FOREACH(stp, hp, ls_hash) {
   1259 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
   1260 			NFSX_STATEIDOTHER))
   1261 			break;
   1262 	}
   1263 
   1264 	/*
   1265 	 * If no state id in list, return NFSERR_BADSTATEID.
   1266 	 */
   1267 	if (stp == LIST_END(hp)) {
   1268 		error = NFSERR_BADSTATEID;
   1269 		goto out;
   1270 	}
   1271 	*stpp = stp;
   1272 
   1273 out:
   1274 	NFSEXITCODE(error);
   1275 	return (error);
   1276 }
   1277 
   1278 /*
   1279  * This function gets an nfsstate structure via owner string.
   1280  */
   1281 static void
   1282 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
   1283     struct nfsstate **stpp)
   1284 {
   1285 	struct nfsstate *stp;
   1286 
   1287 	*stpp = NULL;
   1288 	LIST_FOREACH(stp, hp, ls_list) {
   1289 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
   1290 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
   1291 			*stpp = stp;
   1292 			return;
   1293 		}
   1294 	}
   1295 }
   1296 
   1297 /*
   1298  * Lock control function called to update lock status.
   1299  * Returns 0 upon success, -1 if there is no lock and the flags indicate
   1300  * that one isn't to be created and an NFSERR_xxx for other errors.
   1301  * The structures new_stp and new_lop are passed in as pointers that should
   1302  * be set to NULL if the structure is used and shouldn't be free'd.
   1303  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
   1304  * never used and can safely be allocated on the stack. For all other
   1305  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
   1306  * in case they are used.
   1307  */
   1308 APPLESTATIC int
   1309 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
   1310     struct nfslock **new_lopp, struct nfslockconflict *cfp,
   1311     nfsquad_t clientid, nfsv4stateid_t *stateidp,
   1312     __unused struct nfsexstuff *exp,
   1313     struct nfsrv_descript *nd, NFSPROC_T *p)
   1314 {
   1315 	struct nfslock *lop;
   1316 	struct nfsstate *new_stp = *new_stpp;
   1317 	struct nfslock *new_lop = *new_lopp;
   1318 	struct nfsstate *tstp, *mystp, *nstp;
   1319 	int specialid = 0;
   1320 	struct nfslockfile *lfp;
   1321 	struct nfslock *other_lop = NULL;
   1322 	struct nfsstate *stp, *lckstp = NULL;
   1323 	struct nfsclient *clp = NULL;
   1324 	u_int32_t bits;
   1325 	int error = 0, haslock = 0, ret, reterr;
   1326 	int getlckret, delegation = 0, filestruct_locked;
   1327 	fhandle_t nfh;
   1328 	uint64_t first, end;
   1329 	uint32_t lock_flags;
   1330 
   1331 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
   1332 		/*
   1333 		 * Note the special cases of "all 1s" or "all 0s" stateids and
   1334 		 * let reads with all 1s go ahead.
   1335 		 */
   1336 		if (new_stp->ls_stateid.seqid == 0x0 &&
   1337 		    new_stp->ls_stateid.other[0] == 0x0 &&
   1338 		    new_stp->ls_stateid.other[1] == 0x0 &&
   1339 		    new_stp->ls_stateid.other[2] == 0x0)
   1340 			specialid = 1;
   1341 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
   1342 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
   1343 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
   1344 		    new_stp->ls_stateid.other[2] == 0xffffffff)
   1345 			specialid = 2;
   1346 	}
   1347 
   1348 	/*
   1349 	 * Check for restart conditions (client and server).
   1350 	 */
   1351 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
   1352 	    &new_stp->ls_stateid, specialid);
   1353 	if (error)
   1354 		goto out;
   1355 
   1356 	/*
   1357 	 * Check for state resource limit exceeded.
   1358 	 */
   1359 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
   1360 	    nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
   1361 		error = NFSERR_RESOURCE;
   1362 		goto out;
   1363 	}
   1364 
   1365 	/*
   1366 	 * For the lock case, get another nfslock structure,
   1367 	 * just in case we need it.
   1368 	 * Malloc now, before we start sifting through the linked lists,
   1369 	 * in case we have to wait for memory.
   1370 	 */
   1371 tryagain:
   1372 	if (new_stp->ls_flags & NFSLCK_LOCK)
   1373 		MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
   1374 		    M_NFSDLOCK, M_WAITOK);
   1375 	filestruct_locked = 0;
   1376 	reterr = 0;
   1377 	lfp = NULL;
   1378 
   1379 	/*
   1380 	 * Get the lockfile structure for CFH now, so we can do a sanity
   1381 	 * check against the stateid, before incrementing the seqid#, since
   1382 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
   1383 	 * shouldn't be incremented for this case.
   1384 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
   1385 	 * will be handled later.
   1386 	 * If we are doing Lock/LockU and local locking is enabled, sleep
   1387 	 * lock the nfslockfile structure.
   1388 	 */
   1389 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
   1390 	NFSLOCKSTATE();
   1391 	if (getlckret == 0) {
   1392 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
   1393 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
   1394 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
   1395 			    &lfp, &nfh, 1);
   1396 			if (getlckret == 0)
   1397 				filestruct_locked = 1;
   1398 		} else
   1399 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
   1400 			    &lfp, &nfh, 0);
   1401 	}
   1402 	if (getlckret != 0 && getlckret != -1)
   1403 		reterr = getlckret;
   1404 
   1405 	if (filestruct_locked != 0) {
   1406 		LIST_INIT(&lfp->lf_rollback);
   1407 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
   1408 			/*
   1409 			 * For local locking, do the advisory locking now, so
   1410 			 * that any conflict can be detected. A failure later
   1411 			 * can be rolled back locally. If an error is returned,
   1412 			 * struct nfslockfile has been unlocked and any local
   1413 			 * locking rolled back.
   1414 			 */
   1415 			NFSUNLOCKSTATE();
   1416 			reterr = nfsrv_locallock(vp, lfp,
   1417 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
   1418 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
   1419 			NFSLOCKSTATE();
   1420 		}
   1421 	}
   1422 
   1423 	if (specialid == 0) {
   1424 	    if (new_stp->ls_flags & NFSLCK_TEST) {
   1425 		/*
   1426 		 * RFC 3530 does not list LockT as an op that renews a
   1427 		 * lease, but the concensus seems to be that it is ok
   1428 		 * for a server to do so.
   1429 		 */
   1430 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
   1431 		    (nfsquad_t)((u_quad_t)0), NULL, p);
   1432 
   1433 		/*
   1434 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
   1435 		 * error returns for LockT, just go ahead and test for a lock,
   1436 		 * since there are no locks for this client, but other locks
   1437 		 * can conflict. (ie. same client will always be false)
   1438 		 */
   1439 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
   1440 		    error = 0;
   1441 		lckstp = new_stp;
   1442 	    } else {
   1443 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
   1444 		(nfsquad_t)((u_quad_t)0), NULL, p);
   1445 	      if (error == 0)
   1446 		/*
   1447 		 * Look up the stateid
   1448 		 */
   1449 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
   1450 		  new_stp->ls_flags, &stp);
   1451 	      /*
   1452 	       * do some sanity checks for an unconfirmed open or a
   1453 	       * stateid that refers to the wrong file, for an open stateid
   1454 	       */
   1455 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
   1456 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
   1457 		   (getlckret == 0 && stp->ls_lfp != lfp)))
   1458 			error = NFSERR_BADSTATEID;
   1459 	      if (error == 0 &&
   1460 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
   1461 		  getlckret == 0 && stp->ls_lfp != lfp)
   1462 			error = NFSERR_BADSTATEID;
   1463 
   1464 	      /*
   1465 	       * If the lockowner stateid doesn't refer to the same file,
   1466 	       * I believe that is considered ok, since some clients will
   1467 	       * only create a single lockowner and use that for all locks
   1468 	       * on all files.
   1469 	       * For now, log it as a diagnostic, instead of considering it
   1470 	       * a BadStateid.
   1471 	       */
   1472 	      if (error == 0 && (stp->ls_flags &
   1473 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
   1474 		  getlckret == 0 && stp->ls_lfp != lfp) {
   1475 #ifdef DIAGNOSTIC
   1476 		  printf("Got a lock statid for different file open\n");
   1477 #endif
   1478 		  /*
   1479 		  error = NFSERR_BADSTATEID;
   1480 		  */
   1481 	      }
   1482 
   1483 	      if (error == 0) {
   1484 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
   1485 			/*
   1486 			 * If haslock set, we've already checked the seqid.
   1487 			 */
   1488 			if (!haslock) {
   1489 			    if (stp->ls_flags & NFSLCK_OPEN)
   1490 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
   1491 				    stp->ls_openowner, new_stp->ls_op);
   1492 			    else
   1493 				error = NFSERR_BADSTATEID;
   1494 			}
   1495 			if (!error)
   1496 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
   1497 			if (lckstp)
   1498 			    /*
   1499 			     * I believe this should be an error, but it
   1500 			     * isn't obvious what NFSERR_xxx would be
   1501 			     * appropriate, so I'll use NFSERR_INVAL for now.
   1502 			     */
   1503 			    error = NFSERR_INVAL;
   1504 			else
   1505 			    lckstp = new_stp;
   1506 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
   1507 			/*
   1508 			 * If haslock set, ditto above.
   1509 			 */
   1510 			if (!haslock) {
   1511 			    if (stp->ls_flags & NFSLCK_OPEN)
   1512 				error = NFSERR_BADSTATEID;
   1513 			    else
   1514 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
   1515 				    stp, new_stp->ls_op);
   1516 			}
   1517 			lckstp = stp;
   1518 		    } else {
   1519 			lckstp = stp;
   1520 		    }
   1521 	      }
   1522 	      /*
   1523 	       * If the seqid part of the stateid isn't the same, return
   1524 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
   1525 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
   1526 	       * nfsrv_returnoldstateid is set. (The concensus on the email
   1527 	       * list was that most clients would prefer to not receive
   1528 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
   1529 	       * is what will happen, so I use the nfsrv_returnoldstateid to
   1530 	       * allow for either server configuration.)
   1531 	       */
   1532 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
   1533 		  (!(new_stp->ls_flags & NFSLCK_CHECK) ||
   1534 		   nfsrv_returnoldstateid))
   1535 		    error = NFSERR_OLDSTATEID;
   1536 	    }
   1537 	}
   1538 
   1539 	/*
   1540 	 * Now we can check for grace.
   1541 	 */
   1542 	if (!error)
   1543 		error = nfsrv_checkgrace(new_stp->ls_flags);
   1544 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
   1545 		nfsrv_checkstable(clp))
   1546 		error = NFSERR_NOGRACE;
   1547 	/*
   1548 	 * If we successfully Reclaimed state, note that.
   1549 	 */
   1550 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
   1551 		nfsrv_markstable(clp);
   1552 
   1553 	/*
   1554 	 * At this point, either error == NFSERR_BADSTATEID or the
   1555 	 * seqid# has been updated, so we can return any error.
   1556 	 * If error == 0, there may be an error in:
   1557 	 *    nd_repstat - Set by the calling function.
   1558 	 *    reterr - Set above, if getting the nfslockfile structure
   1559 	 *       or acquiring the local lock failed.
   1560 	 *    (If both of these are set, nd_repstat should probably be
   1561 	 *     returned, since that error was detected before this
   1562 	 *     function call.)
   1563 	 */
   1564 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
   1565 		if (error == 0) {
   1566 			if (nd->nd_repstat != 0)
   1567 				error = nd->nd_repstat;
   1568 			else
   1569 				error = reterr;
   1570 		}
   1571 		if (filestruct_locked != 0) {
   1572 			/* Roll back local locks. */
   1573 			NFSUNLOCKSTATE();
   1574 			nfsrv_locallock_rollback(vp, lfp, p);
   1575 			NFSLOCKSTATE();
   1576 			nfsrv_unlocklf(lfp);
   1577 		}
   1578 		NFSUNLOCKSTATE();
   1579 		goto out;
   1580 	}
   1581 
   1582 	/*
   1583 	 * Check the nfsrv_getlockfile return.
   1584 	 * Returned -1 if no structure found.
   1585 	 */
   1586 	if (getlckret == -1) {
   1587 		error = NFSERR_EXPIRED;
   1588 		/*
   1589 		 * Called from lockt, so no lock is OK.
   1590 		 */
   1591 		if (new_stp->ls_flags & NFSLCK_TEST) {
   1592 			error = 0;
   1593 		} else if (new_stp->ls_flags &
   1594 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
   1595 			/*
   1596 			 * Called to check for a lock, OK if the stateid is all
   1597 			 * 1s or all 0s, but there should be an nfsstate
   1598 			 * otherwise.
   1599 			 * (ie. If there is no open, I'll assume no share
   1600 			 *  deny bits.)
   1601 			 */
   1602 			if (specialid)
   1603 				error = 0;
   1604 			else
   1605 				error = NFSERR_BADSTATEID;
   1606 		}
   1607 		NFSUNLOCKSTATE();
   1608 		goto out;
   1609 	}
   1610 
   1611 	/*
   1612 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
   1613 	 * For NFSLCK_CHECK, allow a read if write access is granted,
   1614 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
   1615 	 * which implies a conflicting deny can't exist.
   1616 	 */
   1617 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
   1618 	    /*
   1619 	     * Four kinds of state id:
   1620 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
   1621 	     * - stateid for an open
   1622 	     * - stateid for a delegation
   1623 	     * - stateid for a lock owner
   1624 	     */
   1625 	    if (!specialid) {
   1626 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
   1627 		    delegation = 1;
   1628 		    mystp = stp;
   1629 		    nfsrv_delaydelegtimeout(stp);
   1630 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
   1631 		    mystp = stp;
   1632 		} else {
   1633 		    mystp = stp->ls_openstp;
   1634 		}
   1635 		/*
   1636 		 * If locking or checking, require correct access
   1637 		 * bit set.
   1638 		 */
   1639 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
   1640 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
   1641 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
   1642 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
   1643 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
   1644 		     !(mystp->ls_flags & NFSLCK_READACCESS)) ||
   1645 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
   1646 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
   1647 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
   1648 			if (filestruct_locked != 0) {
   1649 				/* Roll back local locks. */
   1650 				NFSUNLOCKSTATE();
   1651 				nfsrv_locallock_rollback(vp, lfp, p);
   1652 				NFSLOCKSTATE();
   1653 				nfsrv_unlocklf(lfp);
   1654 			}
   1655 			NFSUNLOCKSTATE();
   1656 			error = NFSERR_OPENMODE;
   1657 			goto out;
   1658 		}
   1659 	    } else
   1660 		mystp = NULL;
   1661 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
   1662 		/*
   1663 		 * Check for a conflicting deny bit.
   1664 		 */
   1665 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
   1666 		    if (tstp != mystp) {
   1667 			bits = tstp->ls_flags;
   1668 			bits >>= NFSLCK_SHIFT;
   1669 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
   1670 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
   1671 				vp, p);
   1672 			    if (ret == 1) {
   1673 				/*
   1674 				* nfsrv_clientconflict unlocks state
   1675 				 * when it returns non-zero.
   1676 				 */
   1677 				lckstp = NULL;
   1678 				goto tryagain;
   1679 			    }
   1680 			    if (ret == 0)
   1681 				NFSUNLOCKSTATE();
   1682 			    if (ret == 2)
   1683 				error = NFSERR_PERM;
   1684 			    else
   1685 				error = NFSERR_OPENMODE;
   1686 			    goto out;
   1687 			}
   1688 		    }
   1689 		}
   1690 
   1691 		/* We're outta here */
   1692 		NFSUNLOCKSTATE();
   1693 		goto out;
   1694 	    }
   1695 	}
   1696 
   1697 	/*
   1698 	 * For setattr, just get rid of all the Delegations for other clients.
   1699 	 */
   1700 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
   1701 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
   1702 		if (ret) {
   1703 			/*
   1704 			 * nfsrv_cleandeleg() unlocks state when it
   1705 			 * returns non-zero.
   1706 			 */
   1707 			if (ret == -1) {
   1708 				lckstp = NULL;
   1709 				goto tryagain;
   1710 			}
   1711 			error = ret;
   1712 			goto out;
   1713 		}
   1714 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
   1715 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
   1716 		     LIST_EMPTY(&lfp->lf_deleg))) {
   1717 			NFSUNLOCKSTATE();
   1718 			goto out;
   1719 		}
   1720 	}
   1721 
   1722 	/*
   1723 	 * Check for a conflicting delegation. If one is found, call
   1724 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
   1725 	 * been set yet, it will get the lock. Otherwise, it will recall
   1726 	 * the delegation. Then, we try try again...
   1727 	 * I currently believe the conflict algorithm to be:
   1728 	 * For Lock Ops (Lock/LockT/LockU)
   1729 	 * - there is a conflict iff a different client has a write delegation
   1730 	 * For Reading (Read Op)
   1731 	 * - there is a conflict iff a different client has a write delegation
   1732 	 *   (the specialids are always a different client)
   1733 	 * For Writing (Write/Setattr of size)
   1734 	 * - there is a conflict if a different client has any delegation
   1735 	 * - there is a conflict if the same client has a read delegation
   1736 	 *   (I don't understand why this isn't allowed, but that seems to be
   1737 	 *    the current concensus?)
   1738 	 */
   1739 	tstp = LIST_FIRST(&lfp->lf_deleg);
   1740 	while (tstp != LIST_END(&lfp->lf_deleg)) {
   1741 	    nstp = LIST_NEXT(tstp, ls_file);
   1742 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
   1743 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
   1744 		  (new_lop->lo_flags & NFSLCK_READ))) &&
   1745 		  clp != tstp->ls_clp &&
   1746 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
   1747 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
   1748 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
   1749 		  (clp != tstp->ls_clp ||
   1750 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
   1751 		if (filestruct_locked != 0) {
   1752 			/* Roll back local locks. */
   1753 			NFSUNLOCKSTATE();
   1754 			nfsrv_locallock_rollback(vp, lfp, p);
   1755 			NFSLOCKSTATE();
   1756 			nfsrv_unlocklf(lfp);
   1757 		}
   1758 		ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
   1759 		if (ret) {
   1760 		    /*
   1761 		     * nfsrv_delegconflict unlocks state when it
   1762 		     * returns non-zero, which it always does.
   1763 		     */
   1764 		    if (other_lop) {
   1765 			FREE((caddr_t)other_lop, M_NFSDLOCK);
   1766 			other_lop = NULL;
   1767 		    }
   1768 		    if (ret == -1) {
   1769 			lckstp = NULL;
   1770 			goto tryagain;
   1771 		    }
   1772 		    error = ret;
   1773 		    goto out;
   1774 		}
   1775 		/* Never gets here. */
   1776 	    }
   1777 	    tstp = nstp;
   1778 	}
   1779 
   1780 	/*
   1781 	 * Handle the unlock case by calling nfsrv_updatelock().
   1782 	 * (Should I have done some access checking above for unlock? For now,
   1783 	 *  just let it happen.)
   1784 	 */
   1785 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
   1786 		first = new_lop->lo_first;
   1787 		end = new_lop->lo_end;
   1788 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
   1789 		stateidp->seqid = ++(stp->ls_stateid.seqid);
   1790 		stateidp->other[0] = stp->ls_stateid.other[0];
   1791 		stateidp->other[1] = stp->ls_stateid.other[1];
   1792 		stateidp->other[2] = stp->ls_stateid.other[2];
   1793 		if (filestruct_locked != 0) {
   1794 			NFSUNLOCKSTATE();
   1795 			/* Update the local locks. */
   1796 			nfsrv_localunlock(vp, lfp, first, end, p);
   1797 			NFSLOCKSTATE();
   1798 			nfsrv_unlocklf(lfp);
   1799 		}
   1800 		NFSUNLOCKSTATE();
   1801 		goto out;
   1802 	}
   1803 
   1804 	/*
   1805 	 * Search for a conflicting lock. A lock conflicts if:
   1806 	 * - the lock range overlaps and
   1807 	 * - at least one lock is a write lock and
   1808 	 * - it is not owned by the same lock owner
   1809 	 */
   1810 	if (!delegation) {
   1811 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
   1812 	    if (new_lop->lo_end > lop->lo_first &&
   1813 		new_lop->lo_first < lop->lo_end &&
   1814 		(new_lop->lo_flags == NFSLCK_WRITE ||
   1815 		 lop->lo_flags == NFSLCK_WRITE) &&
   1816 		lckstp != lop->lo_stp &&
   1817 		(clp != lop->lo_stp->ls_clp ||
   1818 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
   1819 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
   1820 		    lckstp->ls_ownerlen))) {
   1821 		if (other_lop) {
   1822 		    FREE((caddr_t)other_lop, M_NFSDLOCK);
   1823 		    other_lop = NULL;
   1824 		}
   1825 		ret = nfsrv_clientconflict(lop->lo_stp->ls_clp,&haslock,vp,p);
   1826 		if (ret == 1) {
   1827 		    if (filestruct_locked != 0) {
   1828 			/* Roll back local locks. */
   1829 			nfsrv_locallock_rollback(vp, lfp, p);
   1830 			NFSLOCKSTATE();
   1831 			nfsrv_unlocklf(lfp);
   1832 			NFSUNLOCKSTATE();
   1833 		    }
   1834 		    /*
   1835 		     * nfsrv_clientconflict() unlocks state when it
   1836 		     * returns non-zero.
   1837 		     */
   1838 		    lckstp = NULL;
   1839 		    goto tryagain;
   1840 		}
   1841 		/*
   1842 		 * Found a conflicting lock, so record the conflict and
   1843 		 * return the error.
   1844 		 */
   1845 		if (cfp != NULL && ret == 0) {
   1846 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
   1847 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
   1848 		    cfp->cl_first = lop->lo_first;
   1849 		    cfp->cl_end = lop->lo_end;
   1850 		    cfp->cl_flags = lop->lo_flags;
   1851 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
   1852 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
   1853 			cfp->cl_ownerlen);
   1854 		}
   1855 		if (ret == 2)
   1856 		    error = NFSERR_PERM;
   1857 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
   1858 		    error = NFSERR_RECLAIMCONFLICT;
   1859 		else if (new_stp->ls_flags & NFSLCK_CHECK)
   1860 		    error = NFSERR_LOCKED;
   1861 		else
   1862 		    error = NFSERR_DENIED;
   1863 		if (filestruct_locked != 0 && ret == 0) {
   1864 			/* Roll back local locks. */
   1865 			NFSUNLOCKSTATE();
   1866 			nfsrv_locallock_rollback(vp, lfp, p);
   1867 			NFSLOCKSTATE();
   1868 			nfsrv_unlocklf(lfp);
   1869 		}
   1870 		if (ret == 0)
   1871 			NFSUNLOCKSTATE();
   1872 		goto out;
   1873 	    }
   1874 	  }
   1875 	}
   1876 
   1877 	/*
   1878 	 * We only get here if there was no lock that conflicted.
   1879 	 */
   1880 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
   1881 		NFSUNLOCKSTATE();
   1882 		goto out;
   1883 	}
   1884 
   1885 	/*
   1886 	 * We only get here when we are creating or modifying a lock.
   1887 	 * There are two variants:
   1888 	 * - exist_lock_owner where lock_owner exists
   1889 	 * - open_to_lock_owner with new lock_owner
   1890 	 */
   1891 	first = new_lop->lo_first;
   1892 	end = new_lop->lo_end;
   1893 	lock_flags = new_lop->lo_flags;
   1894 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
   1895 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
   1896 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
   1897 		stateidp->other[0] = lckstp->ls_stateid.other[0];
   1898 		stateidp->other[1] = lckstp->ls_stateid.other[1];
   1899 		stateidp->other[2] = lckstp->ls_stateid.other[2];
   1900 	} else {
   1901 		/*
   1902 		 * The new open_to_lock_owner case.
   1903 		 * Link the new nfsstate into the lists.
   1904 		 */
   1905 		new_stp->ls_seq = new_stp->ls_opentolockseq;
   1906 		nfsrvd_refcache(new_stp->ls_op);
   1907 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
   1908 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
   1909 		    clp->lc_clientid.lval[0];
   1910 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
   1911 		    clp->lc_clientid.lval[1];
   1912 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
   1913 		    nfsrv_nextstateindex(clp);
   1914 		new_stp->ls_clp = clp;
   1915 		LIST_INIT(&new_stp->ls_lock);
   1916 		new_stp->ls_openstp = stp;
   1917 		new_stp->ls_lfp = lfp;
   1918 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
   1919 		    lfp);
   1920 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
   1921 		    new_stp, ls_hash);
   1922 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
   1923 		*new_lopp = NULL;
   1924 		*new_stpp = NULL;
   1925 		newnfsstats.srvlockowners++;
   1926 		nfsrv_openpluslock++;
   1927 	}
   1928 	if (filestruct_locked != 0) {
   1929 		NFSUNLOCKSTATE();
   1930 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
   1931 		NFSLOCKSTATE();
   1932 		nfsrv_unlocklf(lfp);
   1933 	}
   1934 	NFSUNLOCKSTATE();
   1935 
   1936 out:
   1937 	if (haslock) {
   1938 		NFSLOCKV4ROOTMUTEX();
   1939 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
   1940 		NFSUNLOCKV4ROOTMUTEX();
   1941 	}
   1942 	if (other_lop)
   1943 		FREE((caddr_t)other_lop, M_NFSDLOCK);
   1944 	NFSEXITCODE2(error, nd);
   1945 	return (error);
   1946 }
   1947 
   1948 /*
   1949  * Check for state errors for Open.
   1950  * repstat is passed back out as an error if more critical errors
   1951  * are not detected.
   1952  */
   1953 APPLESTATIC int
   1954 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
   1955     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
   1956     NFSPROC_T *p, int repstat)
   1957 {
   1958 	struct nfsstate *stp, *nstp;
   1959 	struct nfsclient *clp;
   1960 	struct nfsstate *ownerstp;
   1961 	struct nfslockfile *lfp, *new_lfp;
   1962 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
   1963 
   1964 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
   1965 		readonly = 1;
   1966 	/*
   1967 	 * Check for restart conditions (client and server).
   1968 	 */
   1969 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
   1970 		&new_stp->ls_stateid, 0);
   1971 	if (error)
   1972 		goto out;
   1973 
   1974 	/*
   1975 	 * Check for state resource limit exceeded.
   1976 	 * Technically this should be SMP protected, but the worst
   1977 	 * case error is "out by one or two" on the count when it
   1978 	 * returns NFSERR_RESOURCE and the limit is just a rather
   1979 	 * arbitrary high water mark, so no harm is done.
   1980 	 */
   1981 	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
   1982 		error = NFSERR_RESOURCE;
   1983 		goto out;
   1984 	}
   1985 
   1986 tryagain:
   1987 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
   1988 	    M_NFSDLOCKFILE, M_WAITOK);
   1989 	if (vp)
   1990 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, &new_lfp,
   1991 		    NULL, p);
   1992 	NFSLOCKSTATE();
   1993 	/*
   1994 	 * Get the nfsclient structure.
   1995 	 */
   1996 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
   1997 	    (nfsquad_t)((u_quad_t)0), NULL, p);
   1998 
   1999 	/*
   2000 	 * Look up the open owner. See if it needs confirmation and
   2001 	 * check the seq#, as required.
   2002 	 */
   2003 	if (!error)
   2004 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
   2005 
   2006 	if (!error && ownerstp) {
   2007 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
   2008 		    new_stp->ls_op);
   2009 		/*
   2010 		 * If the OpenOwner hasn't been confirmed, assume the
   2011 		 * old one was a replay and this one is ok.
   2012 		 * See: RFC3530 Sec. 14.2.18.
   2013 		 */
   2014 		if (error == NFSERR_BADSEQID &&
   2015 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
   2016 			error = 0;
   2017 	}
   2018 
   2019 	/*
   2020 	 * Check for grace.
   2021 	 */
   2022 	if (!error)
   2023 		error = nfsrv_checkgrace(new_stp->ls_flags);
   2024 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
   2025 		nfsrv_checkstable(clp))
   2026 		error = NFSERR_NOGRACE;
   2027 
   2028 	/*
   2029 	 * If none of the above errors occurred, let repstat be
   2030 	 * returned.
   2031 	 */
   2032 	if (repstat && !error)
   2033 		error = repstat;
   2034 	if (error) {
   2035 		NFSUNLOCKSTATE();
   2036 		if (haslock) {
   2037 			NFSLOCKV4ROOTMUTEX();
   2038 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
   2039 			NFSUNLOCKV4ROOTMUTEX();
   2040 		}
   2041 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
   2042 		goto out;
   2043 	}
   2044 
   2045 	/*
   2046 	 * If vp == NULL, the file doesn't exist yet, so return ok.
   2047 	 * (This always happens on the first pass, so haslock must be 0.)
   2048 	 */
   2049 	if (vp == NULL) {
   2050 		NFSUNLOCKSTATE();
   2051 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
   2052 		goto out;
   2053 	}
   2054 
   2055 	/*
   2056 	 * Get the structure for the underlying file.
   2057 	 */
   2058 	if (getfhret)
   2059 		error = getfhret;
   2060 	else
   2061 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
   2062 		    NULL, 0);
   2063 	if (new_lfp)
   2064 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
   2065 	if (error) {
   2066 		NFSUNLOCKSTATE();
   2067 		if (haslock) {
   2068 			NFSLOCKV4ROOTMUTEX();
   2069 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
   2070 			NFSUNLOCKV4ROOTMUTEX();
   2071 		}
   2072 		goto out;
   2073 	}
   2074 
   2075 	/*
   2076 	 * Search for a conflicting open/share.
   2077 	 */
   2078 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
   2079 	    /*
   2080 	     * For Delegate_Cur, search for the matching Delegation,
   2081 	     * which indicates no conflict.
   2082 	     * An old delegation should have been recovered by the
   2083 	     * client doing a Claim_DELEGATE_Prev, so I won't let
   2084 	     * it match and return NFSERR_EXPIRED. Should I let it
   2085 	     * match?
   2086 	     */
   2087 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
   2088 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
   2089 		    stateidp->seqid == stp->ls_stateid.seqid &&
   2090 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
   2091 			  NFSX_STATEIDOTHER))
   2092 			break;
   2093 	    }
   2094 	    if (stp == LIST_END(&lfp->lf_deleg) ||
   2095 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
   2096 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
   2097 		NFSUNLOCKSTATE();
   2098 		if (haslock) {
   2099 			NFSLOCKV4ROOTMUTEX();
   2100 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
   2101 			NFSUNLOCKV4ROOTMUTEX();
   2102 		}
   2103 		error = NFSERR_EXPIRED;
   2104 		goto out;
   2105 	    }
   2106 	}
   2107 
   2108 	/*
   2109 	 * Check for access/deny bit conflicts. I check for the same
   2110 	 * owner as well, in case the client didn't bother.
   2111 	 */
   2112 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
   2113 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
   2114 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
   2115 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
   2116 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
   2117 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
   2118 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
   2119 			if (ret == 1) {
   2120 				/*
   2121 				 * nfsrv_clientconflict() unlocks
   2122 				 * state when it returns non-zero.
   2123 				 */
   2124 				goto tryagain;
   2125 			}
   2126 			if (ret == 2)
   2127 				error = NFSERR_PERM;
   2128 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
   2129 				error = NFSERR_RECLAIMCONFLICT;
   2130 			else
   2131 				error = NFSERR_SHAREDENIED;
   2132 			if (ret == 0)
   2133 				NFSUNLOCKSTATE();
   2134 			if (haslock) {
   2135 				NFSLOCKV4ROOTMUTEX();
   2136 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
   2137 				NFSUNLOCKV4ROOTMUTEX();
   2138 			}
   2139 			goto out;
   2140 		}
   2141 	}
   2142 
   2143 	/*
   2144 	 * Check for a conflicting delegation. If one is found, call
   2145 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
   2146 	 * been set yet, it will get the lock. Otherwise, it will recall
   2147 	 * the delegation. Then, we try try again...
   2148 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
   2149 	 *  isn't a conflict.)
   2150 	 * I currently believe the conflict algorithm to be:
   2151 	 * For Open with Read Access and Deny None
   2152 	 * - there is a conflict iff a different client has a write delegation
   2153 	 * For Open with other Write Access or any Deny except None
   2154 	 * - there is a conflict if a different client has any delegation
   2155 	 * - there is a conflict if the same client has a read delegation
   2156 	 *   (The current concensus is that this last case should be
   2157 	 *    considered a conflict since the client with a read delegation
   2158 	 *    could have done an Open with ReadAccess and WriteDeny
   2159 	 *    locally and then not have checked for the WriteDeny.)
   2160 	 * Don't check for a Reclaim, since that will be dealt with
   2161 	 * by nfsrv_openctrl().
   2162 	 */
   2163 	if (!(new_stp->ls_flags &
   2164 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
   2165 	    stp = LIST_FIRST(&lfp->lf_deleg);
   2166 	    while (stp != LIST_END(&lfp->lf_deleg)) {
   2167 		nstp = LIST_NEXT(stp, ls_file);
   2168 		if ((readonly && stp->ls_clp != clp &&
   2169 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
   2170 		    (!readonly && (stp->ls_clp != clp ||
   2171 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
   2172 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
   2173 			if (ret) {
   2174 			    /*
   2175 			     * nfsrv_delegconflict() unlocks state
   2176 			     * when it returns non-zero.
   2177 			     */
   2178 			    if (ret == -1)
   2179 				goto tryagain;
   2180 			    error = ret;
   2181 			    goto out;
   2182 			}
   2183 		}
   2184 		stp = nstp;
   2185 	    }
   2186 	}
   2187 	NFSUNLOCKSTATE();
   2188 	if (haslock) {
   2189 		NFSLOCKV4ROOTMUTEX();
   2190 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
   2191 		NFSUNLOCKV4ROOTMUTEX();
   2192 	}
   2193 
   2194 out:
   2195 	NFSEXITCODE2(error, nd);
   2196 	return (error);
   2197 }
   2198 
   2199 /*
   2200  * Open control function to create/update open state for an open.
   2201  */
   2202 APPLESTATIC int
   2203 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
   2204     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
   2205     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
   2206     NFSPROC_T *p, u_quad_t filerev)
   2207 {
   2208 	struct nfsstate *new_stp = *new_stpp;
   2209 	struct nfsstate *stp, *nstp;
   2210 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
   2211 	struct nfslockfile *lfp, *new_lfp;
   2212 	struct nfsclient *clp;
   2213 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
   2214 	int readonly = 0, cbret = 1, getfhret = 0;
   2215 
   2216 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
   2217 		readonly = 1;
   2218 	/*
   2219 	 * Check for restart conditions (client and server).
   2220 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
   2221 	 * If an error does show up, return NFSERR_EXPIRED, since the
   2222 	 * the seqid# has already been incremented.
   2223 	 */
   2224 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
   2225 	    &new_stp->ls_stateid, 0);
   2226 	if (error) {
   2227 		printf("Nfsd: openctrl unexpected restart err=%d\n",
   2228 		    error);
   2229 		error = NFSERR_EXPIRED;
   2230 		goto out;
   2231 	}
   2232 
   2233 tryagain:
   2234 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
   2235 	    M_NFSDLOCKFILE, M_WAITOK);
   2236 	MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
   2237 	    M_NFSDSTATE, M_WAITOK);
   2238 	MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
   2239 	    M_NFSDSTATE, M_WAITOK);
   2240 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, &new_lfp,
   2241 	    NULL, p);
   2242 	NFSLOCKSTATE();
   2243 	/*
   2244 	 * Get the client structure. Since the linked lists could be changed
   2245 	 * by other nfsd processes if this process does a tsleep(), one of
   2246 	 * two things must be done.
   2247 	 * 1 - don't tsleep()
   2248 	 * or
   2249 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
   2250 	 *     before using the lists, since this lock stops the other
   2251 	 *     nfsd. This should only be used for rare cases, since it
   2252 	 *     essentially single threads the nfsd.
   2253 	 *     At this time, it is only done for cases where the stable
   2254 	 *     storage file must be written prior to completion of state
   2255 	 *     expiration.
   2256 	 */
   2257 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
   2258 	    (nfsquad_t)((u_quad_t)0), NULL, p);
   2259 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
   2260 	    clp->lc_program) {
   2261 		/*
   2262 		 * This happens on the first open for a client
   2263 		 * that supports callbacks.
   2264 		 */
   2265 		NFSUNLOCKSTATE();
   2266 		/*
   2267 		 * Although nfsrv_docallback() will sleep, clp won't
   2268 		 * go away, since they are only removed when the
   2269 		 * nfsv4_lock() has blocked the nfsd threads. The
   2270 		 * fields in clp can change, but having multiple
   2271 		 * threads do this Null callback RPC should be
   2272 		 * harmless.
   2273 		 */
   2274 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
   2275 		    NULL, 0, NULL, NULL, NULL, p);
   2276 		NFSLOCKSTATE();
   2277 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
   2278 		if (!cbret)
   2279 			clp->lc_flags |= LCL_CALLBACKSON;
   2280 	}
   2281 
   2282 	/*
   2283 	 * Look up the open owner. See if it needs confirmation and
   2284 	 * check the seq#, as required.
   2285 	 */
   2286 	if (!error)
   2287 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
   2288 
   2289 	if (error) {
   2290 		NFSUNLOCKSTATE();
   2291 		printf("Nfsd: openctrl unexpected state err=%d\n",
   2292 			error);
   2293 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
   2294 		free((caddr_t)new_open, M_NFSDSTATE);
   2295 		free((caddr_t)new_deleg, M_NFSDSTATE);
   2296 		if (haslock) {
   2297 			NFSLOCKV4ROOTMUTEX();
   2298 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
   2299 			NFSUNLOCKV4ROOTMUTEX();
   2300 		}
   2301 		error = NFSERR_EXPIRED;
   2302 		goto out;
   2303 	}
   2304 
   2305 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
   2306 		nfsrv_markstable(clp);
   2307 
   2308 	/*
   2309 	 * Get the structure for the underlying file.
   2310 	 */
   2311 	if (getfhret)
   2312 		error = getfhret;
   2313 	else
   2314 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
   2315 		    NULL, 0);
   2316 	if (new_lfp)
   2317 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
   2318 	if (error) {
   2319 		NFSUNLOCKSTATE();
   2320 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
   2321 		    error);
   2322 		free((caddr_t)new_open, M_NFSDSTATE);
   2323 		free((caddr_t)new_deleg, M_NFSDSTATE);
   2324 		if (haslock) {
   2325 			NFSLOCKV4ROOTMUTEX();
   2326 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
   2327 			NFSUNLOCKV4ROOTMUTEX();
   2328 		}
   2329 		goto out;
   2330 	}
   2331 
   2332 	/*
   2333 	 * Search for a conflicting open/share.
   2334 	 */
   2335 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
   2336 	    /*
   2337 	     * For Delegate_Cur, search for the matching Delegation,
   2338 	     * which indicates no conflict.
   2339 	     * An old delegation should have been recovered by the
   2340 	     * client doing a Claim_DELEGATE_Prev, so I won't let
   2341 	     * it match and return NFSERR_EXPIRED. Should I let it
   2342 	     * match?
   2343 	     */
   2344 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
   2345 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
   2346 		    stateidp->seqid == stp->ls_stateid.seqid &&
   2347 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
   2348 			NFSX_STATEIDOTHER))
   2349 			break;
   2350 	    }
   2351 	    if (stp == LIST_END(&lfp->lf_deleg) ||
   2352 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
   2353 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
   2354 		NFSUNLOCKSTATE();
   2355 		printf("Nfsd openctrl unexpected expiry\n");
   2356 		free((caddr_t)new_open, M_NFSDSTATE);
   2357 		free((caddr_t)new_deleg, M_NFSDSTATE);
   2358 		if (haslock) {
   2359 			NFSLOCKV4ROOTMUTEX();
   2360 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
   2361 			NFSUNLOCKV4ROOTMUTEX();
   2362 		}
   2363 		error = NFSERR_EXPIRED;
   2364 		goto out;
   2365 	    }
   2366 
   2367 	    /*
   2368 	     * Don't issue a Delegation, since one already exists and
   2369 	     * delay delegation timeout, as required.
   2370 	     */
   2371 	    delegate = 0;
   2372 	    nfsrv_delaydelegtimeout(stp);
   2373 	}
   2374 
   2375 	/*
   2376 	 * Check for access/deny bit conflicts. I also check for the
   2377 	 * same owner, since the client might not have bothered to check.
   2378 	 * Also, note an open for the same file and owner, if found,
   2379 	 * which is all we do here for Delegate_Cur, since conflict
   2380 	 * checking is already done.
   2381 	 */
   2382 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
   2383 		if (ownerstp && stp->ls_openowner == ownerstp)
   2384 			openstp = stp;
   2385 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
   2386 		    /*
   2387 		     * If another client has the file open, the only
   2388 		     * delegation that can be issued is a Read delegation
   2389 		     * and only if it is a Read open with Deny none.
   2390 		     */
   2391 		    if (clp != stp->ls_clp) {
   2392 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
   2393 			    NFSLCK_READACCESS)
   2394 			    writedeleg = 0;
   2395 			else
   2396 			    delegate = 0;
   2397 		    }
   2398 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
   2399 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
   2400 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
   2401 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
   2402 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
   2403 			if (ret == 1) {
   2404 				/*
   2405 				 * nfsrv_clientconflict() unlocks state
   2406 				 * when it returns non-zero.
   2407 				 */
   2408 				free((caddr_t)new_open, M_NFSDSTATE);
   2409 				free((caddr_t)new_deleg, M_NFSDSTATE);
   2410 				openstp = NULL;
   2411 				goto tryagain;
   2412 			}
   2413 			if (ret == 2)
   2414 				error = NFSERR_PERM;
   2415 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
   2416 				error = NFSERR_RECLAIMCONFLICT;
   2417 			else
   2418 				error = NFSERR_SHAREDENIED;
   2419 			if (ret == 0)
   2420 				NFSUNLOCKSTATE();
   2421 			if (haslock) {
   2422 				NFSLOCKV4ROOTMUTEX();
   2423 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
   2424 				NFSUNLOCKV4ROOTMUTEX();
   2425 			}
   2426 			free((caddr_t)new_open, M_NFSDSTATE);
   2427 			free((caddr_t)new_deleg, M_NFSDSTATE);
   2428 			printf("nfsd openctrl unexpected client cnfl\n");
   2429 			goto out;
   2430 		    }
   2431 		}
   2432 	}
   2433 
   2434 	/*
   2435 	 * Check for a conflicting delegation. If one is found, call
   2436 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
   2437 	 * been set yet, it will get the lock. Otherwise, it will recall
   2438 	 * the delegation. Then, we try try again...
   2439 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
   2440 	 *  isn't a conflict.)
   2441 	 * I currently believe the conflict algorithm to be:
   2442 	 * For Open with Read Access and Deny None
   2443 	 * - there is a conflict iff a different client has a write delegation
   2444 	 * For Open with other Write Access or any Deny except None
   2445 	 * - there is a conflict if a different client has any delegation
   2446 	 * - there is a conflict if the same client has a read delegation
   2447 	 *   (The current concensus is that this last case should be
   2448 	 *    considered a conflict since the client with a read delegation
   2449 	 *    could have done an Open with ReadAccess and WriteDeny
   2450 	 *    locally and then not have checked for the WriteDeny.)
   2451 	 */
   2452 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
   2453 	    stp = LIST_FIRST(&lfp->lf_deleg);
   2454 	    while (stp != LIST_END(&lfp->lf_deleg)) {
   2455 		nstp = LIST_NEXT(stp, ls_file);
   2456 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
   2457 			writedeleg = 0;
   2458 		else
   2459 			delegate = 0;
   2460 		if ((readonly && stp->ls_clp != clp &&
   2461 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
   2462 		    (!readonly && (stp->ls_clp != clp ||
   2463 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
   2464 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
   2465 			delegate = 2;
   2466 		    } else {
   2467 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
   2468 			if (ret) {
   2469 			    /*
   2470 			     * nfsrv_delegconflict() unlocks state
   2471 			     * when it returns non-zero.
   2472 			     */
   2473 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
   2474 			    free((caddr_t)new_open, M_NFSDSTATE);
   2475 			    free((caddr_t)new_deleg, M_NFSDSTATE);
   2476 			    if (ret == -1) {
   2477 				openstp = NULL;
   2478 				goto tryagain;
   2479 			    }
   2480 			    error = ret;
   2481 			    goto out;
   2482 			}
   2483 		    }
   2484 		}
   2485 		stp = nstp;
   2486 	    }
   2487 	}
   2488 
   2489 	/*
   2490 	 * We only get here if there was no open that conflicted.
   2491 	 * If an open for the owner exists, or in the access/deny bits.
   2492 	 * Otherwise it is a new open. If the open_owner hasn't been
   2493 	 * confirmed, replace the open with the new one needing confirmation,
   2494 	 * otherwise add the open.
   2495 	 */
   2496 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
   2497 	    /*
   2498 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
   2499 	     * a match. If found, just move the old delegation to the current
   2500 	     * delegation list and issue open. If not found, return
   2501 	     * NFSERR_EXPIRED.
   2502 	     */
   2503 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
   2504 		if (stp->ls_lfp == lfp) {
   2505 		    /* Found it */
   2506 		    if (stp->ls_clp != clp)
   2507 			panic("olddeleg clp");
   2508 		    LIST_REMOVE(stp, ls_list);
   2509 		    LIST_REMOVE(stp, ls_hash);
   2510 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
   2511 		    stp->ls_stateid.seqid = delegstateidp->seqid = 0;
   2512 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
   2513 			clp->lc_clientid.lval[0];
   2514 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
   2515 			clp->lc_clientid.lval[1];
   2516 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
   2517 			nfsrv_nextstateindex(clp);
   2518 		    stp->ls_compref = nd->nd_compref;
   2519 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
   2520 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
   2521 			stp->ls_stateid), stp, ls_hash);
   2522 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
   2523 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
   2524 		    else
   2525 			*rflagsp |= NFSV4OPEN_READDELEGATE;
   2526 		    clp->lc_delegtime = NFSD_MONOSEC +
   2527 			nfsrv_lease + NFSRV_LEASEDELTA;
   2528 
   2529 		    /*
   2530 		     * Now, do the associated open.
   2531 		     */
   2532 		    new_open->ls_stateid.seqid = 0;
   2533 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
   2534 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
   2535 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
   2536 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
   2537 			NFSLCK_OPEN;
   2538 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
   2539 			new_open->ls_flags |= (NFSLCK_READACCESS |
   2540 			    NFSLCK_WRITEACCESS);
   2541 		    else
   2542 			new_open->ls_flags |= NFSLCK_READACCESS;
   2543 		    new_open->ls_uid = new_stp->ls_uid;
   2544 		    new_open->ls_lfp = lfp;
   2545 		    new_open->ls_clp = clp;
   2546 		    LIST_INIT(&new_open->ls_open);
   2547 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
   2548 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
   2549 			new_open, ls_hash);
   2550 		    /*
   2551 		     * and handle the open owner
   2552 		     */
   2553 		    if (ownerstp) {
   2554 			new_open->ls_openowner = ownerstp;
   2555 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
   2556 		    } else {
   2557 			new_open->ls_openowner = new_stp;
   2558 			new_stp->ls_flags = 0;
   2559 			nfsrvd_refcache(new_stp->ls_op);
   2560 			new_stp->ls_noopens = 0;
   2561 			LIST_INIT(&new_stp->ls_open);
   2562 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
   2563 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
   2564 			*new_stpp = NULL;
   2565 			newnfsstats.srvopenowners++;
   2566 			nfsrv_openpluslock++;
   2567 		    }
   2568 		    openstp = new_open;
   2569 		    new_open = NULL;
   2570 		    newnfsstats.srvopens++;
   2571 		    nfsrv_openpluslock++;
   2572 		    break;
   2573 		}
   2574 	    }
   2575 	    if (stp == LIST_END(&clp->lc_olddeleg))
   2576 		error = NFSERR_EXPIRED;
   2577 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
   2578 	    /*
   2579 	     * Scan to see that no delegation for this client and file
   2580 	     * doesn't already exist.
   2581 	     * There also shouldn't yet be an Open for this file and
   2582 	     * openowner.
   2583 	     */
   2584 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
   2585 		if (stp->ls_clp == clp)
   2586 		    break;
   2587 	    }
   2588 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
   2589 		/*
   2590 		 * This is the Claim_Previous case with a delegation
   2591 		 * type != Delegate_None.
   2592 		 */
   2593 		/*
   2594 		 * First, add the delegation. (Although we must issue the
   2595 		 * delegation, we can also ask for an immediate return.)
   2596 		 */
   2597 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
   2598 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
   2599 		    clp->lc_clientid.lval[0];
   2600 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
   2601 		    clp->lc_clientid.lval[1];
   2602 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
   2603 		    nfsrv_nextstateindex(clp);
   2604 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
   2605 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
   2606 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
   2607 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
   2608 		} else {
   2609 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
   2610 			NFSLCK_READACCESS);
   2611 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
   2612 		}
   2613 		new_deleg->ls_uid = new_stp->ls_uid;
   2614 		new_deleg->ls_lfp = lfp;
   2615 		new_deleg->ls_clp = clp;
   2616 		new_deleg->ls_filerev = filerev;
   2617 		new_deleg->ls_compref = nd->nd_compref;
   2618 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
   2619 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
   2620 		    new_deleg->ls_stateid), new_deleg, ls_hash);
   2621 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
   2622 		new_deleg = NULL;
   2623 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
   2624 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
   2625 		     LCL_CALLBACKSON ||
   2626 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
   2627 		    !NFSVNO_DELEGOK(vp))
   2628 		    *rflagsp |= NFSV4OPEN_RECALL;
   2629 		newnfsstats.srvdelegates++;
   2630 		nfsrv_openpluslock++;
   2631 		nfsrv_delegatecnt++;
   2632 
   2633 		/*
   2634 		 * Now, do the associated open.
   2635 		 */
   2636 		new_open->ls_stateid.seqid = 0;
   2637 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
   2638 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
   2639 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
   2640 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
   2641 		    NFSLCK_OPEN;
   2642 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
   2643 			new_open->ls_flags |= (NFSLCK_READACCESS |
   2644 			    NFSLCK_WRITEACCESS);
   2645 		else
   2646 			new_open->ls_flags |= NFSLCK_READACCESS;
   2647 		new_open->ls_uid = new_stp->ls_uid;
   2648 		new_open->ls_lfp = lfp;
   2649 		new_open->ls_clp = clp;
   2650 		LIST_INIT(&new_open->ls_open);
   2651 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
   2652 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
   2653 		   new_open, ls_hash);
   2654 		/*
   2655 		 * and handle the open owner
   2656 		 */
   2657 		if (ownerstp) {
   2658 		    new_open->ls_openowner = ownerstp;
   2659 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
   2660 		} else {
   2661 		    new_open->ls_openowner = new_stp;
   2662 		    new_stp->ls_flags = 0;
   2663 		    nfsrvd_refcache(new_stp->ls_op);
   2664 		    new_stp->ls_noopens = 0;
   2665 		    LIST_INIT(&new_stp->ls_open);
   2666 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
   2667 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
   2668 		    *new_stpp = NULL;
   2669 		    newnfsstats.srvopenowners++;
   2670 		    nfsrv_openpluslock++;
   2671 		}
   2672 		openstp = new_open;
   2673 		new_open = NULL;
   2674 		newnfsstats.srvopens++;
   2675 		nfsrv_openpluslock++;
   2676 	    } else {
   2677 		error = NFSERR_RECLAIMCONFLICT;
   2678 	    }
   2679 	} else if (ownerstp) {
   2680 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
   2681 		    /* Replace the open */
   2682 		    if (ownerstp->ls_op)
   2683 			nfsrvd_derefcache(ownerstp->ls_op);
   2684 		    ownerstp->ls_op = new_stp->ls_op;
   2685 		    nfsrvd_refcache(ownerstp->ls_op);
   2686 		    ownerstp->ls_seq = new_stp->ls_seq;
   2687 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
   2688 		    stp = LIST_FIRST(&ownerstp->ls_open);
   2689 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
   2690 			NFSLCK_OPEN;
   2691 		    stp->ls_stateid.seqid = 0;
   2692 		    stp->ls_uid = new_stp->ls_uid;
   2693 		    if (lfp != stp->ls_lfp) {
   2694 			LIST_REMOVE(stp, ls_file);
   2695 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
   2696 			stp->ls_lfp = lfp;
   2697 		    }
   2698 		    openstp = stp;
   2699 		} else if (openstp) {
   2700 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
   2701 		    openstp->ls_stateid.seqid++;
   2702 
   2703 		    /*
   2704 		     * This is where we can choose to issue a delegation.
   2705 		     */
   2706 		    if (delegate && nfsrv_issuedelegs &&
   2707 			writedeleg && !NFSVNO_EXRDONLY(exp) &&
   2708 			(nfsrv_writedelegifpos || !readonly) &&
   2709 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
   2710 			 LCL_CALLBACKSON &&
   2711 			!NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
   2712 			NFSVNO_DELEGOK(vp)) {
   2713 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
   2714 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
   2715 			    = clp->lc_clientid.lval[0];
   2716 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
   2717 			    = clp->lc_clientid.lval[1];
   2718 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
   2719 			    = nfsrv_nextstateindex(clp);
   2720 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
   2721 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
   2722 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
   2723 			new_deleg->ls_uid = new_stp->ls_uid;
   2724 			new_deleg->ls_lfp = lfp;
   2725 			new_deleg->ls_clp = clp;
   2726 			new_deleg->ls_filerev = filerev;
   2727 			new_deleg->ls_compref = nd->nd_compref;
   2728 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
   2729 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
   2730 			    new_deleg->ls_stateid), new_deleg, ls_hash);
   2731 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
   2732 			new_deleg = NULL;
   2733 			newnfsstats.srvdelegates++;
   2734 			nfsrv_openpluslock++;
   2735 			nfsrv_delegatecnt++;
   2736 		    }
   2737 		} else {
   2738 		    new_open->ls_stateid.seqid = 0;
   2739 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
   2740 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
   2741 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
   2742 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
   2743 			NFSLCK_OPEN;
   2744 		    new_open->ls_uid = new_stp->ls_uid;
   2745 		    new_open->ls_openowner = ownerstp;
   2746 		    new_open->ls_lfp = lfp;
   2747 		    new_open->ls_clp = clp;
   2748 		    LIST_INIT(&new_open->ls_open);
   2749 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
   2750 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
   2751 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
   2752 			new_open, ls_hash);
   2753 		    openstp = new_open;
   2754 		    new_open = NULL;
   2755 		    newnfsstats.srvopens++;
   2756 		    nfsrv_openpluslock++;
   2757 
   2758 		    /*
   2759 		     * This is where we can choose to issue a delegation.
   2760 		     */
   2761 		    if (delegate && nfsrv_issuedelegs &&
   2762 			(writedeleg || readonly) &&
   2763 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
   2764 			 LCL_CALLBACKSON &&
   2765 			!NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
   2766 			NFSVNO_DELEGOK(vp)) {
   2767 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
   2768 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
   2769 			    = clp->lc_clientid.lval[0];
   2770 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
   2771 			    = clp->lc_clientid.lval[1];
   2772 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
   2773 			    = nfsrv_nextstateindex(clp);
   2774 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
   2775 			    (nfsrv_writedelegifpos || !readonly)) {
   2776 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
   2777 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
   2778 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
   2779 			} else {
   2780 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
   2781 				NFSLCK_READACCESS);
   2782 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
   2783 			}
   2784 			new_deleg->ls_uid = new_stp->ls_uid;
   2785 			new_deleg->ls_lfp = lfp;
   2786 			new_deleg->ls_clp = clp;
   2787 			new_deleg->ls_filerev = filerev;
   2788 			new_deleg->ls_compref = nd->nd_compref;
   2789 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
   2790 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
   2791 			    new_deleg->ls_stateid), new_deleg, ls_hash);
   2792 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
   2793 			new_deleg = NULL;
   2794 			newnfsstats.srvdelegates++;
   2795 			nfsrv_openpluslock++;
   2796 			nfsrv_delegatecnt++;
   2797 		    }
   2798 		}
   2799 	} else {
   2800 		/*
   2801 		 * New owner case. Start the open_owner sequence with a
   2802 		 * Needs confirmation (unless a reclaim) and hang the
   2803 		 * new open off it.
   2804 		 */
   2805 		new_open->ls_stateid.seqid = 0;
   2806 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
   2807 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
   2808 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
   2809 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
   2810 		    NFSLCK_OPEN;
   2811 		new_open->ls_uid = new_stp->ls_uid;
   2812 		LIST_INIT(&new_open->ls_open);
   2813 		new_open->ls_openowner = new_stp;
   2814 		new_open->ls_lfp = lfp;
   2815 		new_open->ls_clp = clp;
   2816 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
   2817 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
   2818 			new_stp->ls_flags = 0;
   2819 		} else {
   2820 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
   2821 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
   2822 		}
   2823 		nfsrvd_refcache(new_stp->ls_op);
   2824 		new_stp->ls_noopens = 0;
   2825 		LIST_INIT(&new_stp->ls_open);
   2826 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
   2827 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
   2828 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
   2829 		    new_open, ls_hash);
   2830 		openstp = new_open;
   2831 		new_open = NULL;
   2832 		*new_stpp = NULL;
   2833 		newnfsstats.srvopens++;
   2834 		nfsrv_openpluslock++;
   2835 		newnfsstats.srvopenowners++;
   2836 		nfsrv_openpluslock++;
   2837 	}
   2838 	if (!error) {
   2839 		stateidp->seqid = openstp->ls_stateid.seqid;
   2840 		stateidp->other[0] = openstp->ls_stateid.other[0];
   2841 		stateidp->other[1] = openstp->ls_stateid.other[1];
   2842 		stateidp->other[2] = openstp->ls_stateid.other[2];
   2843 	}
   2844 	NFSUNLOCKSTATE();
   2845 	if (haslock) {
   2846 		NFSLOCKV4ROOTMUTEX();
   2847 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
   2848 		NFSUNLOCKV4ROOTMUTEX();
   2849 	}
   2850 	if (new_open)
   2851 		FREE((caddr_t)new_open, M_NFSDSTATE);
   2852 	if (new_deleg)
   2853 		FREE((caddr_t)new_deleg, M_NFSDSTATE);
   2854 
   2855 out:
   2856 	NFSEXITCODE2(error, nd);
   2857 	return (error);
   2858 }
   2859 
   2860 /*
   2861  * Open update. Does the confirm, downgrade and close.
   2862  */
   2863 APPLESTATIC int
   2864 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
   2865     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
   2866 {
   2867 	struct nfsstate *stp, *ownerstp;
   2868 	struct nfsclient *clp;
   2869 	struct nfslockfile *lfp;
   2870 	u_int32_t bits;
   2871 	int error = 0, gotstate = 0, len = 0;
   2872 	u_char client[NFSV4_OPAQUELIMIT];
   2873 
   2874 	/*
   2875 	 * Check for restart conditions (client and server).
   2876 	 */
   2877 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
   2878 	    &new_stp->ls_stateid, 0);
   2879 	if (error)
   2880 		goto out;
   2881 
   2882 	NFSLOCKSTATE();
   2883 	/*
   2884 	 * Get the open structure via clientid and stateid.
   2885 	 */
   2886 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
   2887 	    (nfsquad_t)((u_quad_t)0), NULL, p);
   2888 	if (!error)
   2889 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
   2890 		    new_stp->ls_flags, &stp);
   2891 
   2892 	/*
   2893 	 * Sanity check the open.
   2894 	 */
   2895 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
   2896 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
   2897 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
   2898 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
   2899 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
   2900 		error = NFSERR_BADSTATEID;
   2901 
   2902 	if (!error)
   2903 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
   2904 		    stp->ls_openowner, new_stp->ls_op);
   2905 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
   2906 	    !(new_stp->ls_flags & NFSLCK_CONFIRM))
   2907 		error = NFSERR_OLDSTATEID;
   2908 	if (!error && vnode_vtype(vp) != VREG) {
   2909 		if (vnode_vtype(vp) == VDIR)
   2910 			error = NFSERR_ISDIR;
   2911 		else
   2912 			error = NFSERR_INVAL;
   2913 	}
   2914 
   2915 	if (error) {
   2916 		/*
   2917 		 * If a client tries to confirm an Open with a bad
   2918 		 * seqid# and there are no byte range locks or other Opens
   2919 		 * on the openowner, just throw it away, so the next use of the
   2920 		 * openowner will start a fresh seq#.
   2921 		 */
   2922 		if (error == NFSERR_BADSEQID &&
   2923 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
   2924 		    nfsrv_nootherstate(stp))
   2925 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
   2926 		NFSUNLOCKSTATE();
   2927 		goto out;
   2928 	}
   2929 
   2930 	/*
   2931 	 * Set the return stateid.
   2932 	 */
   2933 	stateidp->seqid = stp->ls_stateid.seqid + 1;
   2934 	stateidp->other[0] = stp->ls_stateid.other[0];
   2935 	stateidp->other[1] = stp->ls_stateid.other[1];
   2936 	stateidp->other[2] = stp->ls_stateid.other[2];
   2937 	/*
   2938 	 * Now, handle the three cases.
   2939 	 */
   2940 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
   2941 		/*
   2942 		 * If the open doesn't need confirmation, it seems to me that
   2943 		 * there is a client error, but I'll just log it and keep going?
   2944 		 */
   2945 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
   2946 			printf("Nfsv4d: stray open confirm\n");
   2947 		stp->ls_openowner->ls_flags = 0;
   2948 		stp->ls_stateid.seqid++;
   2949 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
   2950 			clp->lc_flags |= LCL_STAMPEDSTABLE;
   2951 			len = clp->lc_idlen;
   2952 			NFSBCOPY(clp->lc_id, client, len);
   2953 			gotstate = 1;
   2954 		}
   2955 		NFSUNLOCKSTATE();
   2956 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
   2957 		ownerstp = stp->ls_openowner;
   2958 		lfp = stp->ls_lfp;
   2959 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
   2960 			/* Get the lf lock */
   2961 			nfsrv_locklf(lfp);
   2962 			NFSUNLOCKSTATE();
   2963 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
   2964 				NFSLOCKSTATE();
   2965 				nfsrv_unlocklf(lfp);
   2966 				NFSUNLOCKSTATE();
   2967 			}
   2968 		} else {
   2969 			(void) nfsrv_freeopen(stp, NULL, 0, p);
   2970 			NFSUNLOCKSTATE();
   2971 		}
   2972 	} else {
   2973 		/*
   2974 		 * Update the share bits, making sure that the new set are a
   2975 		 * subset of the old ones.
   2976 		 */
   2977 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
   2978 		if (~(stp->ls_flags) & bits) {
   2979 			NFSUNLOCKSTATE();
   2980 			error = NFSERR_INVAL;
   2981 			goto out;
   2982 		}
   2983 		stp->ls_flags = (bits | NFSLCK_OPEN);
   2984 		stp->ls_stateid.seqid++;
   2985 		NFSUNLOCKSTATE();
   2986 	}
   2987 
   2988 	/*
   2989 	 * If the client just confirmed its first open, write a timestamp
   2990 	 * to the stable storage file.
   2991 	 */
   2992 	if (gotstate != 0) {
   2993 		nfsrv_writestable(client, len, NFSNST_NEWSTATE, p);
   2994 		nfsrv_backupstable();
   2995 	}
   2996 
   2997 out:
   2998 	NFSEXITCODE2(error, nd);
   2999 	return (error);
   3000 }
   3001 
   3002 /*
   3003  * Delegation update. Does the purge and return.
   3004  */
   3005 APPLESTATIC int
   3006 nfsrv_delegupdate(nfsquad_t clientid, nfsv4stateid_t *stateidp,
   3007     vnode_t vp, int op, struct ucred *cred, NFSPROC_T *p)
   3008 {
   3009 	struct nfsstate *stp;
   3010 	struct nfsclient *clp;
   3011 	int error = 0;
   3012 	fhandle_t fh;
   3013 
   3014 	/*
   3015 	 * Do a sanity check against the file handle for DelegReturn.
   3016 	 */
   3017 	if (vp) {
   3018 		error = nfsvno_getfh(vp, &fh, p);
   3019 		if (error)
   3020 			goto out;
   3021 	}
   3022 	/*
   3023 	 * Check for restart conditions (client and server).
   3024 	 */
   3025 	if (op == NFSV4OP_DELEGRETURN)
   3026 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
   3027 			stateidp, 0);
   3028 	else
   3029 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
   3030 			stateidp, 0);
   3031 
   3032 	NFSLOCKSTATE();
   3033 	/*
   3034 	 * Get the open structure via clientid and stateid.
   3035 	 */
   3036 	if (!error)
   3037 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
   3038 		(nfsquad_t)((u_quad_t)0), NULL, p);
   3039 	if (error) {
   3040 		if (error == NFSERR_CBPATHDOWN)
   3041 			error = 0;
   3042 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
   3043 			error = NFSERR_STALESTATEID;
   3044 	}
   3045 	if (!error && op == NFSV4OP_DELEGRETURN) {
   3046 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
   3047 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid)
   3048 		error = NFSERR_OLDSTATEID;
   3049 	}
   3050 	/*
   3051 	 * NFSERR_EXPIRED means that the state has gone away,
   3052 	 * so Delegations have been purged. Just return ok.
   3053 	 */
   3054 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
   3055 		NFSUNLOCKSTATE();
   3056 		error = 0;
   3057 		goto out;
   3058 	}
   3059 	if (error) {
   3060 		NFSUNLOCKSTATE();
   3061 		goto out;
   3062 	}
   3063 
   3064 	if (op == NFSV4OP_DELEGRETURN) {
   3065 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
   3066 		    sizeof (fhandle_t))) {
   3067 			NFSUNLOCKSTATE();
   3068 			error = NFSERR_BADSTATEID;
   3069 			goto out;
   3070 		}
   3071 		nfsrv_freedeleg(stp);
   3072 	} else {
   3073 		nfsrv_freedeleglist(&clp->lc_olddeleg);
   3074 	}
   3075 	NFSUNLOCKSTATE();
   3076 	error = 0;
   3077 
   3078 out:
   3079 	NFSEXITCODE(error);
   3080 	return (error);
   3081 }
   3082 
   3083 /*
   3084  * Release lock owner.
   3085  */
   3086 APPLESTATIC int
   3087 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
   3088     NFSPROC_T *p)
   3089 {
   3090 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
   3091 	struct nfsclient *clp;
   3092 	int error = 0;
   3093 
   3094 	/*
   3095 	 * Check for restart conditions (client and server).
   3096 	 */
   3097 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
   3098 	    &new_stp->ls_stateid, 0);
   3099 	if (error)
   3100 		goto out;
   3101 
   3102 	NFSLOCKSTATE();
   3103 	/*
   3104 	 * Get the lock owner by name.
   3105 	 */
   3106 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
   3107 	    (nfsquad_t)((u_quad_t)0), NULL, p);
   3108 	if (error) {
   3109 		NFSUNLOCKSTATE();
   3110 		goto out;
   3111 	}
   3112 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
   3113 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
   3114 		stp = LIST_FIRST(&openstp->ls_open);
   3115 		while (stp != LIST_END(&openstp->ls_open)) {
   3116 		    nstp = LIST_NEXT(stp, ls_list);
   3117 		    /*
   3118 		     * If the owner matches, check for locks and
   3119 		     * then free or return an error.
   3120 		     */
   3121 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
   3122 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
   3123 			 stp->ls_ownerlen)){
   3124 			if (LIST_EMPTY(&stp->ls_lock)) {
   3125 			    nfsrv_freelockowner(stp, NULL, 0, p);
   3126 			} else {
   3127 			    NFSUNLOCKSTATE();
   3128 			    error = NFSERR_LOCKSHELD;
   3129 			    goto out;
   3130 			}
   3131 		    }
   3132 		    stp = nstp;
   3133 		}
   3134 	    }
   3135 	}
   3136 	NFSUNLOCKSTATE();
   3137 
   3138 out:
   3139 	NFSEXITCODE(error);
   3140 	return (error);
   3141 }
   3142 
   3143 /*
   3144  * Get the file handle for a lock structure.
   3145  */
   3146 static int
   3147 nfsrv_getlockfh(vnode_t vp, u_short flags,
   3148     struct nfslockfile **new_lfpp, fhandle_t *nfhp, NFSPROC_T *p)
   3149 {
   3150 	fhandle_t *fhp = NULL;
   3151 	struct nfslockfile *new_lfp;
   3152 	int error;
   3153 
   3154 	/*
   3155 	 * For lock, use the new nfslock structure, otherwise just
   3156 	 * a fhandle_t on the stack.
   3157 	 */
   3158 	if (flags & NFSLCK_OPEN) {
   3159 		new_lfp = *new_lfpp;
   3160 		fhp = &new_lfp->lf_fh;
   3161 	} else if (nfhp) {
   3162 		fhp = nfhp;
   3163 	} else {
   3164 		panic("nfsrv_getlockfh");
   3165 	}
   3166 	error = nfsvno_getfh(vp, fhp, p);
   3167 	NFSEXITCODE(error);
   3168 	return (error);
   3169 }
   3170 
   3171 /*
   3172  * Get an nfs lock structure. Allocate one, as required, and return a
   3173  * pointer to it.
   3174  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
   3175  */
   3176 static int
   3177 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
   3178     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
   3179 {
   3180 	struct nfslockfile *lfp;
   3181 	fhandle_t *fhp = NULL, *tfhp;
   3182 	struct nfslockhashhead *hp;
   3183 	struct nfslockfile *new_lfp = NULL;
   3184 
   3185 	/*
   3186 	 * For lock, use the new nfslock structure, otherwise just
   3187 	 * a fhandle_t on the stack.
   3188 	 */
   3189 	if (flags & NFSLCK_OPEN) {
   3190 		new_lfp = *new_lfpp;
   3191 		fhp = &new_lfp->lf_fh;
   3192 	} else if (nfhp) {
   3193 		fhp = nfhp;
   3194 	} else {
   3195 		panic("nfsrv_getlockfile");
   3196 	}
   3197 
   3198 	hp = NFSLOCKHASH(fhp);
   3199 	LIST_FOREACH(lfp, hp, lf_hash) {
   3200 		tfhp = &lfp->lf_fh;
   3201 		if (NFSVNO_CMPFH(fhp, tfhp)) {
   3202 			if (lockit)
   3203 				nfsrv_locklf(lfp);
   3204 			*lfpp = lfp;
   3205 			return (0);
   3206 		}
   3207 	}
   3208 	if (!(flags & NFSLCK_OPEN))
   3209 		return (-1);
   3210 
   3211 	/*
   3212 	 * No match, so chain the new one into the list.
   3213 	 */
   3214 	LIST_INIT(&new_lfp->lf_open);
   3215 	LIST_INIT(&new_lfp->lf_lock);
   3216 	LIST_INIT(&new_lfp->lf_deleg);
   3217 	LIST_INIT(&new_lfp->lf_locallock);
   3218 	LIST_INIT(&new_lfp->lf_rollback);
   3219 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
   3220 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
   3221 	new_lfp->lf_usecount = 0;
   3222 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
   3223 	*lfpp = new_lfp;
   3224 	*new_lfpp = NULL;
   3225 	return (0);
   3226 }
   3227 
   3228 /*
   3229  * This function adds a nfslock lock structure to the list for the associated
   3230  * nfsstate and nfslockfile structures. It will be inserted after the
   3231  * entry pointed at by insert_lop.
   3232  */
   3233 static void
   3234 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
   3235     struct nfsstate *stp, struct nfslockfile *lfp)
   3236 {
   3237 	struct nfslock *lop, *nlop;
   3238 
   3239 	new_lop->lo_stp = stp;
   3240 	new_lop->lo_lfp = lfp;
   3241 
   3242 	if (stp != NULL) {
   3243 		/* Insert in increasing lo_first order */
   3244 		lop = LIST_FIRST(&lfp->lf_lock);
   3245 		if (lop == LIST_END(&lfp->lf_lock) ||
   3246 		    new_lop->lo_first <= lop->lo_first) {
   3247 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
   3248 		} else {
   3249 			nlop = LIST_NEXT(lop, lo_lckfile);
   3250 			while (nlop != LIST_END(&lfp->lf_lock) &&
   3251 			       nlop->lo_first < new_lop->lo_first) {
   3252 				lop = nlop;
   3253 				nlop = LIST_NEXT(lop, lo_lckfile);
   3254 			}
   3255 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
   3256 		}
   3257 	} else {
   3258 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
   3259 	}
   3260 
   3261 	/*
   3262 	 * Insert after insert_lop, which is overloaded as stp or lfp for
   3263 	 * an empty list.
   3264 	 */
   3265 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
   3266 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
   3267 	else if ((struct nfsstate *)insert_lop == stp)
   3268 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
   3269 	else
   3270 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
   3271 	if (stp != NULL) {
   3272 		newnfsstats.srvlocks++;
   3273 		nfsrv_openpluslock++;
   3274 	}
   3275 }
   3276 
   3277 /*
   3278  * This function updates the locking for a lock owner and given file. It
   3279  * maintains a list of lock ranges ordered on increasing file offset that
   3280  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
   3281  * It always adds new_lop to the list and sometimes uses the one pointed
   3282  * at by other_lopp.
   3283  */
   3284 static void
   3285 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
   3286     struct nfslock **other_lopp, struct nfslockfile *lfp)
   3287 {
   3288 	struct nfslock *new_lop = *new_lopp;
   3289 	struct nfslock *lop, *tlop, *ilop;
   3290 	struct nfslock *other_lop = *other_lopp;
   3291 	int unlock = 0, myfile = 0;
   3292 	u_int64_t tmp;
   3293 
   3294 	/*
   3295 	 * Work down the list until the lock is merged.
   3296 	 */
   3297 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
   3298 		unlock = 1;
   3299 	if (stp != NULL) {
   3300 		ilop = (struct nfslock *)stp;
   3301 		lop = LIST_FIRST(&stp->ls_lock);
   3302 	} else {
   3303 		ilop = (struct nfslock *)lfp;
   3304 		lop = LIST_FIRST(&lfp->lf_locallock);
   3305 	}
   3306 	while (lop != NULL) {
   3307 	    /*
   3308 	     * Only check locks for this file that aren't before the start of
   3309 	     * new lock's range.
   3310 	     */
   3311 	    if (lop->lo_lfp == lfp) {
   3312 	      myfile = 1;
   3313 	      if (lop->lo_end >= new_lop->lo_first) {
   3314 		if (new_lop->lo_end < lop->lo_first) {
   3315 			/*
   3316 			 * If the new lock ends before the start of the
   3317 			 * current lock's range, no merge, just insert
   3318 			 * the new lock.
   3319 			 */
   3320 			break;
   3321 		}
   3322 		if (new_lop->lo_flags == lop->lo_flags ||
   3323 		    (new_lop->lo_first <= lop->lo_first &&
   3324 		     new_lop->lo_end >= lop->lo_end)) {
   3325 			/*
   3326 			 * This lock can be absorbed by the new lock/unlock.
   3327 			 * This happens when it covers the entire range
   3328 			 * of the old lock or is contiguous
   3329 			 * with the old lock and is of the same type or an
   3330 			 * unlock.
   3331 			 */
   3332 			if (lop->lo_first < new_lop->lo_first)
   3333 				new_lop->lo_first = lop->lo_first;
   3334 			if (lop->lo_end > new_lop->lo_end)
   3335 				new_lop->lo_end = lop->lo_end;
   3336 			tlop = lop;
   3337 			lop = LIST_NEXT(lop, lo_lckowner);
   3338 			nfsrv_freenfslock(tlop);
   3339 			continue;
   3340 		}
   3341 
   3342 		/*
   3343 		 * All these cases are for contiguous locks that are not the
   3344 		 * same type, so they can't be merged.
   3345 		 */
   3346 		if (new_lop->lo_first <= lop->lo_first) {
   3347 			/*
   3348 			 * This case is where the new lock overlaps with the
   3349 			 * first part of the old lock. Move the start of the
   3350 			 * old lock to just past the end of the new lock. The
   3351 			 * new lock will be inserted in front of the old, since
   3352 			 * ilop hasn't been updated. (We are done now.)
   3353 			 */
   3354 			lop->lo_first = new_lop->lo_end;
   3355 			break;
   3356 		}
   3357 		if (new_lop->lo_end >= lop->lo_end) {
   3358 			/*
   3359 			 * This case is where the new lock overlaps with the
   3360 			 * end of the old lock's range. Move the old lock's
   3361 			 * end to just before the new lock's first and insert
   3362 			 * the new lock after the old lock.
   3363 			 * Might not be done yet, since the new lock could
   3364 			 * overlap further locks with higher ranges.
   3365 			 */
   3366 			lop->lo_end = new_lop->lo_first;
   3367 			ilop = lop;
   3368 			lop = LIST_NEXT(lop, lo_lckowner);
   3369 			continue;
   3370 		}
   3371 		/*
   3372 		 * The final case is where the new lock's range is in the
   3373 		 * middle of the current lock's and splits the current lock
   3374 		 * up. Use *other_lopp to handle the second part of the
   3375 		 * split old lock range. (We are done now.)
   3376 		 * For unlock, we use new_lop as other_lop and tmp, since
   3377 		 * other_lop and new_lop are the same for this case.
   3378 		 * We noted the unlock case above, so we don't need
   3379 		 * new_lop->lo_flags any longer.
   3380 		 */
   3381 		tmp = new_lop->lo_first;
   3382 		if (other_lop == NULL) {
   3383 			if (!unlock)
   3384 				panic("nfsd srv update unlock");
   3385 			other_lop = new_lop;
   3386 			*new_lopp = NULL;
   3387 		}
   3388 		other_lop->lo_first = new_lop->lo_end;
   3389 		other_lop->lo_end = lop->lo_end;
   3390 		other_lop->lo_flags = lop->lo_flags;
   3391 		other_lop->lo_stp = stp;
   3392 		other_lop->lo_lfp = lfp;
   3393 		lop->lo_end = tmp;
   3394 		nfsrv_insertlock(other_lop, lop, stp, lfp);
   3395 		*other_lopp = NULL;
   3396 		ilop = lop;
   3397 		break;
   3398 	      }
   3399 	    }
   3400 	    ilop = lop;
   3401 	    lop = LIST_NEXT(lop, lo_lckowner);
   3402 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
   3403 		break;
   3404 	}
   3405 
   3406 	/*
   3407 	 * Insert the new lock in the list at the appropriate place.
   3408 	 */
   3409 	if (!unlock) {
   3410 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
   3411 		*new_lopp = NULL;
   3412 	}
   3413 }
   3414 
   3415 /*
   3416  * This function handles sequencing of locks, etc.
   3417  * It returns an error that indicates what the caller should do.
   3418  */
   3419 static int
   3420 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
   3421     struct nfsstate *stp, struct nfsrvcache *op)
   3422 {
   3423 	int error = 0;
   3424 
   3425 	if (op != nd->nd_rp)
   3426 		panic("nfsrvstate checkseqid");
   3427 	if (!(op->rc_flag & RC_INPROG))
   3428 		panic("nfsrvstate not inprog");
   3429 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
   3430 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
   3431 		panic("nfsrvstate op refcnt");
   3432 	}
   3433 	if ((stp->ls_seq + 1) == seqid) {
   3434 		if (stp->ls_op)
   3435 			nfsrvd_derefcache(stp->ls_op);
   3436 		stp->ls_op = op;
   3437 		nfsrvd_refcache(op);
   3438 		stp->ls_seq = seqid;
   3439 		goto out;
   3440 	} else if (stp->ls_seq == seqid && stp->ls_op &&
   3441 		op->rc_xid == stp->ls_op->rc_xid &&
   3442 		op->rc_refcnt == 0 &&
   3443 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
   3444 		op->rc_cksum == stp->ls_op->rc_cksum) {
   3445 		if (stp->ls_op->rc_flag & RC_INPROG) {
   3446 			error = NFSERR_DONTREPLY;
   3447 			goto out;
   3448 		}
   3449 		nd->nd_rp = stp->ls_op;
   3450 		nd->nd_rp->rc_flag |= RC_INPROG;
   3451 		nfsrvd_delcache(op);
   3452 		error = NFSERR_REPLYFROMCACHE;
   3453 		goto out;
   3454 	}
   3455 	error = NFSERR_BADSEQID;
   3456 
   3457 out:
   3458 	NFSEXITCODE2(error, nd);
   3459 	return (error);
   3460 }
   3461 
   3462 /*
   3463  * Get the client ip address for callbacks. If the strings can't be parsed,
   3464  * just set lc_program to 0 to indicate no callbacks are possible.
   3465  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
   3466  *  the address to the client's transport address. This won't be used
   3467  *  for callbacks, but can be printed out by newnfsstats for info.)
   3468  * Return error if the xdr can't be parsed, 0 otherwise.
   3469  */
   3470 APPLESTATIC int
   3471 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
   3472 {
   3473 	u_int32_t *tl;
   3474 	u_char *cp, *cp2;
   3475 	int i, j;
   3476 	struct sockaddr_in *rad, *sad;
   3477 	u_char protocol[5], addr[24];
   3478 	int error = 0, cantparse = 0;
   3479 	union {
   3480 		u_long ival;
   3481 		u_char cval[4];
   3482 	} ip;
   3483 	union {
   3484 		u_short sval;
   3485 		u_char cval[2];
   3486 	} port;
   3487 
   3488 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
   3489 	rad->sin_family = AF_INET;
   3490 	rad->sin_len = sizeof (struct sockaddr_in);
   3491 	rad->sin_addr.s_addr = 0;
   3492 	rad->sin_port = 0;
   3493 	clp->lc_req.nr_client = NULL;
   3494 	clp->lc_req.nr_lock = 0;
   3495 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
   3496 	i = fxdr_unsigned(int, *tl);
   3497 	if (i >= 3 && i <= 4) {
   3498 		error = nfsrv_mtostr(nd, protocol, i);
   3499 		if (error)
   3500 			goto nfsmout;
   3501 		if (!strcmp(protocol, "tcp")) {
   3502 			clp->lc_flags |= LCL_TCPCALLBACK;
   3503 			clp->lc_req.nr_sotype = SOCK_STREAM;
   3504 			clp->lc_req.nr_soproto = IPPROTO_TCP;
   3505 		} else if (!strcmp(protocol, "udp")) {
   3506 			clp->lc_req.nr_sotype = SOCK_DGRAM;
   3507 			clp->lc_req.nr_soproto = IPPROTO_UDP;
   3508 		} else {
   3509 			cantparse = 1;
   3510 		}
   3511 	} else {
   3512 		cantparse = 1;
   3513 		if (i > 0) {
   3514 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
   3515 			if (error)
   3516 				goto nfsmout;
   3517 		}
   3518 	}
   3519 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
   3520 	i = fxdr_unsigned(int, *tl);
   3521 	if (i < 0) {
   3522 		error = NFSERR_BADXDR;
   3523 		goto nfsmout;
   3524 	} else if (i == 0) {
   3525 		cantparse = 1;
   3526 	} else if (!cantparse && i <= 23 && i >= 11) {
   3527 		error = nfsrv_mtostr(nd, addr, i);
   3528 		if (error)
   3529 			goto nfsmout;
   3530 
   3531 		/*
   3532 		 * Parse out the address fields. We expect 6 decimal numbers
   3533 		 * separated by '.'s.
   3534 		 */
   3535 		cp = addr;
   3536 		i = 0;
   3537 		while (*cp && i < 6) {
   3538 			cp2 = cp;
   3539 			while (*cp2 && *cp2 != '.')
   3540 				cp2++;
   3541 			if (*cp2)
   3542 				*cp2++ = '\0';
   3543 			else if (i != 5) {
   3544 				cantparse = 1;
   3545 				break;
   3546 			}
   3547 			j = nfsrv_getipnumber(cp);
   3548 			if (j >= 0) {
   3549 				if (i < 4)
   3550 					ip.cval[3 - i] = j;
   3551 				else
   3552 					port.cval[5 - i] = j;
   3553 			} else {
   3554 				cantparse = 1;
   3555 				break;
   3556 			}
   3557 			cp = cp2;
   3558 			i++;
   3559 		}
   3560 		if (!cantparse) {
   3561 			if (ip.ival != 0x0) {
   3562 				rad->sin_addr.s_addr = htonl(ip.ival);
   3563 				rad->sin_port = htons(port.sval);
   3564 			} else {
   3565 				cantparse = 1;
   3566 			}
   3567 		}
   3568 	} else {
   3569 		cantparse = 1;
   3570 		if (i > 0) {
   3571 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
   3572 			if (error)
   3573 				goto nfsmout;
   3574 		}
   3575 	}
   3576 	if (cantparse) {
   3577 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
   3578 		rad->sin_addr.s_addr = sad->sin_addr.s_addr;
   3579 		rad->sin_port = 0x0;
   3580 		clp->lc_program = 0;
   3581 	}
   3582 nfsmout:
   3583 	NFSEXITCODE2(error, nd);
   3584 	return (error);
   3585 }
   3586 
   3587 /*
   3588  * Turn a string of up to three decimal digits into a number. Return -1 upon
   3589  * error.
   3590  */
   3591 static int
   3592 nfsrv_getipnumber(u_char *cp)
   3593 {
   3594 	int i = 0, j = 0;
   3595 
   3596 	while (*cp) {
   3597 		if (j > 2 || *cp < '0' || *cp > '9')
   3598 			return (-1);
   3599 		i *= 10;
   3600 		i += (*cp - '0');
   3601 		cp++;
   3602 		j++;
   3603 	}
   3604 	if (i < 256)
   3605 		return (i);
   3606 	return (-1);
   3607 }
   3608 
   3609 /*
   3610  * This function checks for restart conditions.
   3611  */
   3612 static int
   3613 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
   3614     nfsv4stateid_t *stateidp, int specialid)
   3615 {
   3616 	int ret = 0;
   3617 
   3618 	/*
   3619 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
   3620 	 * and DelegPurge have a clientid, the rest a stateid.
   3621 	 */
   3622 	if (flags &
   3623 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
   3624 		if (clientid.lval[0] != nfsrvboottime) {
   3625 			ret = NFSERR_STALECLIENTID;
   3626 			goto out;
   3627 		}
   3628 	} else if (stateidp->other[0] != nfsrvboottime &&
   3629 		specialid == 0) {
   3630 		ret = NFSERR_STALESTATEID;
   3631 		goto out;
   3632 	}
   3633 
   3634 	/*
   3635 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
   3636 	 * not use a lock/open owner seqid#, so the check can be done now.
   3637 	 * (The others will be checked, as required, later.)
   3638 	 */
   3639 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
   3640 		goto out;
   3641 
   3642 	NFSLOCKSTATE();
   3643 	ret = nfsrv_checkgrace(flags);
   3644 	NFSUNLOCKSTATE();
   3645 
   3646 out:
   3647 	NFSEXITCODE(ret);
   3648 	return (ret);
   3649 }
   3650 
   3651 /*
   3652  * Check for grace.
   3653  */
   3654 static int
   3655 nfsrv_checkgrace(u_int32_t flags)
   3656 {
   3657 	int error = 0;
   3658 
   3659 	if (nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) {
   3660 		if (flags & NFSLCK_RECLAIM) {
   3661 			error = NFSERR_NOGRACE;
   3662 			goto out;
   3663 		}
   3664 	} else {
   3665 		if (!(flags & NFSLCK_RECLAIM)) {
   3666 			error = NFSERR_GRACE;
   3667 			goto out;
   3668 		}
   3669 
   3670 		/*
   3671 		 * If grace is almost over and we are still getting Reclaims,
   3672 		 * extend grace a bit.
   3673 		 */
   3674 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
   3675 		    nfsrv_stablefirst.nsf_eograce)
   3676 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
   3677 				NFSRV_LEASEDELTA;
   3678 	}
   3679 
   3680 out:
   3681 	NFSEXITCODE(error);
   3682 	return (error);
   3683 }
   3684 
   3685 /*
   3686  * Do a server callback.
   3687  */
   3688 static int
   3689 nfsrv_docallback(struct nfsclient *clp, int procnum,
   3690     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
   3691     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
   3692 {
   3693 	mbuf_t m;
   3694 	u_int32_t *tl;
   3695 	struct nfsrv_descript nfsd, *nd = &nfsd;
   3696 	struct ucred *cred;
   3697 	int error = 0;
   3698 	u_int32_t callback;
   3699 
   3700 	cred = newnfs_getcred();
   3701 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
   3702 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
   3703 		NFSUNLOCKSTATE();
   3704 		panic("docallb");
   3705 	}
   3706 	clp->lc_cbref++;
   3707 
   3708 	/*
   3709 	 * Fill the callback program# and version into the request
   3710 	 * structure for newnfs_connect() to use.
   3711 	 */
   3712 	clp->lc_req.nr_prog = clp->lc_program;
   3713 	clp->lc_req.nr_vers = NFSV4_CBVERS;
   3714 
   3715 	/*
   3716 	 * First, fill in some of the fields of nd and cr.
   3717 	 */
   3718 	nd->nd_flag = ND_NFSV4;
   3719 	if (clp->lc_flags & LCL_GSS)
   3720 		nd->nd_flag |= ND_KERBV;
   3721 	nd->nd_repstat = 0;
   3722 	cred->cr_uid = clp->lc_uid;
   3723 	cred->cr_gid = clp->lc_gid;
   3724 	callback = clp->lc_callback;
   3725 	NFSUNLOCKSTATE();
   3726 	cred->cr_ngroups = 1;
   3727 
   3728 	/*
   3729 	 * Get the first mbuf for the request.
   3730 	 */
   3731 	MGET(m, M_WAITOK, MT_DATA);
   3732 	mbuf_setlen(m, 0);
   3733 	nd->nd_mreq = nd->nd_mb = m;
   3734 	nd->nd_bpos = NFSMTOD(m, caddr_t);
   3735 
   3736 	/*
   3737 	 * and build the callback request.
   3738 	 */
   3739 	if (procnum == NFSV4OP_CBGETATTR) {
   3740 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
   3741 		(void) nfsm_strtom(nd, "CB Getattr", 10);
   3742 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
   3743 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
   3744 		*tl++ = txdr_unsigned(callback);
   3745 		*tl++ = txdr_unsigned(1);
   3746 		*tl = txdr_unsigned(NFSV4OP_CBGETATTR);
   3747 		(void) nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
   3748 		(void) nfsrv_putattrbit(nd, attrbitp);
   3749 	} else if (procnum == NFSV4OP_CBRECALL) {
   3750 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
   3751 		(void) nfsm_strtom(nd, "CB Recall", 9);
   3752 		NFSM_BUILD(tl, u_int32_t *, 5 * NFSX_UNSIGNED + NFSX_STATEID);
   3753 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
   3754 		*tl++ = txdr_unsigned(callback);
   3755 		*tl++ = txdr_unsigned(1);
   3756 		*tl++ = txdr_unsigned(NFSV4OP_CBRECALL);
   3757 		*tl++ = txdr_unsigned(stateidp->seqid);
   3758 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
   3759 		    NFSX_STATEIDOTHER);
   3760 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
   3761 		if (trunc)
   3762 			*tl = newnfs_true;
   3763 		else
   3764 			*tl = newnfs_false;
   3765 		(void) nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
   3766 	} else {
   3767 		nd->nd_procnum = NFSV4PROC_CBNULL;
   3768 	}
   3769 
   3770 	/*
   3771 	 * Call newnfs_connect(), as required, and then newnfs_request().
   3772 	 */
   3773 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
   3774 	if (clp->lc_req.nr_client == NULL) {
   3775 		if (nd->nd_procnum == NFSV4PROC_CBNULL)
   3776 			error = newnfs_connect(NULL, &clp->lc_req, cred,
   3777 			    NULL, 1);
   3778 		else
   3779 			error = newnfs_connect(NULL, &clp->lc_req, cred,
   3780 			    NULL, 3);
   3781 	}
   3782 	newnfs_sndunlock(&clp->lc_req.nr_lock);
   3783 	if (!error) {
   3784 		error = newnfs_request(nd, NULL, clp, &clp->lc_req, NULL,
   3785 		    NULL, cred, clp->lc_program, NFSV4_CBVERS, NULL, 1, NULL,
   3786 		    NULL);
   3787 	}
   3788 	NFSFREECRED(cred);
   3789 
   3790 	/*
   3791 	 * If error is set here, the Callback path isn't working
   3792 	 * properly, so twiddle the appropriate LCL_ flags.
   3793 	 * (nd_repstat != 0 indicates the Callback path is working,
   3794 	 *  but the callback failed on the client.)
   3795 	 */
   3796 	if (error) {
   3797 		/*
   3798 		 * Mark the callback pathway down, which disabled issuing
   3799 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
   3800 		 */
   3801 		NFSLOCKSTATE();
   3802 		clp->lc_flags |= LCL_CBDOWN;
   3803 		NFSUNLOCKSTATE();
   3804 	} else {
   3805 		/*
   3806 		 * Callback worked. If the callback path was down, disable
   3807 		 * callbacks, so no more delegations will be issued. (This
   3808 		 * is done on the assumption that the callback pathway is
   3809 		 * flakey.)
   3810 		 */
   3811 		NFSLOCKSTATE();
   3812 		if (clp->lc_flags & LCL_CBDOWN)
   3813 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
   3814 		NFSUNLOCKSTATE();
   3815 		if (nd->nd_repstat)
   3816 			error = nd->nd_repstat;
   3817 		else if (procnum == NFSV4OP_CBGETATTR)
   3818 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
   3819 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
   3820 			    p, NULL);
   3821 		mbuf_freem(nd->nd_mrep);
   3822 	}
   3823 	NFSLOCKSTATE();
   3824 	clp->lc_cbref--;
   3825 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
   3826 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
   3827 		wakeup(clp);
   3828 	}
   3829 	NFSUNLOCKSTATE();
   3830 
   3831 	NFSEXITCODE(error);
   3832 	return (error);
   3833 }
   3834 
   3835 /*
   3836  * Return the next index# for a clientid. Mostly just increment and return
   3837  * the next one, but... if the 32bit unsigned does actually wrap around,
   3838  * it should be rebooted.
   3839  * At an average rate of one new client per second, it will wrap around in
   3840  * approximately 136 years. (I think the server will have been shut
   3841  * down or rebooted before then.)
   3842  */
   3843 static u_int32_t
   3844 nfsrv_nextclientindex(void)
   3845 {
   3846 	static u_int32_t client_index = 0;
   3847 
   3848 	client_index++;
   3849 	if (client_index != 0)
   3850 		return (client_index);
   3851 
   3852 	printf("%s: out of clientids\n", __func__);
   3853 	return (client_index);
   3854 }
   3855 
   3856 /*
   3857  * Return the next index# for a stateid. Mostly just increment and return
   3858  * the next one, but... if the 32bit unsigned does actually wrap around
   3859  * (will a BSD server stay up that long?), find
   3860  * new start and end values.
   3861  */
   3862 static u_int32_t
   3863 nfsrv_nextstateindex(struct nfsclient *clp)
   3864 {
   3865 	struct nfsstate *stp;
   3866 	int i;
   3867 	u_int32_t canuse, min_index, max_index;
   3868 
   3869 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
   3870 		clp->lc_stateindex++;
   3871 		if (clp->lc_stateindex != clp->lc_statemaxindex)
   3872 			return (clp->lc_stateindex);
   3873 	}
   3874 
   3875 	/*
   3876 	 * Yuck, we've hit the end.
   3877 	 * Look for a new min and max.
   3878 	 */
   3879 	min_index = 0;
   3880 	max_index = 0xffffffff;
   3881 	for (i = 0; i < NFSSTATEHASHSIZE; i++) {
   3882 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
   3883 		if (stp->ls_stateid.other[2] > 0x80000000) {
   3884 		    if (stp->ls_stateid.other[2] < max_index)
   3885 			max_index = stp->ls_stateid.other[2];
   3886 		} else {
   3887 		    if (stp->ls_stateid.other[2] > min_index)
   3888 			min_index = stp->ls_stateid.other[2];
   3889 		}
   3890 	    }
   3891 	}
   3892 
   3893 	/*
   3894 	 * Yikes, highly unlikely, but I'll handle it anyhow.
   3895 	 */
   3896 	if (min_index == 0x80000000 && max_index == 0x80000001) {
   3897 	    canuse = 0;
   3898 	    /*
   3899 	     * Loop around until we find an unused entry. Return that
   3900 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
   3901 	     * (This is one of those rare cases where a goto is the
   3902 	     *  cleanest way to code the loop.)
   3903 	     */
   3904 tryagain:
   3905 	    for (i = 0; i < NFSSTATEHASHSIZE; i++) {
   3906 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
   3907 		    if (stp->ls_stateid.other[2] == canuse) {
   3908 			canuse++;
   3909 			goto tryagain;
   3910 		    }
   3911 		}
   3912 	    }
   3913 	    clp->lc_flags |= LCL_INDEXNOTOK;
   3914 	    return (canuse);
   3915 	}
   3916 
   3917 	/*
   3918 	 * Ok to start again from min + 1.
   3919 	 */
   3920 	clp->lc_stateindex = min_index + 1;
   3921 	clp->lc_statemaxindex = max_index;
   3922 	clp->lc_flags &= ~LCL_INDEXNOTOK;
   3923 	return (clp->lc_stateindex);
   3924 }
   3925 
   3926 /*
   3927  * The following functions handle the stable storage file that deals with
   3928  * the edge conditions described in RFC3530 Sec. 8.6.3.
   3929  * The file is as follows:
   3930  * - a single record at the beginning that has the lease time of the
   3931  *   previous server instance (before the last reboot) and the nfsrvboottime
   3932  *   values for the previous server boots.
   3933  *   These previous boot times are used to ensure that the current
   3934  *   nfsrvboottime does not, somehow, get set to a previous one.
   3935  *   (This is important so that Stale ClientIDs and StateIDs can
   3936  *    be recognized.)
   3937  *   The number of previous nfsvrboottime values preceeds the list.
   3938  * - followed by some number of appended records with:
   3939  *   - client id string
   3940  *   - flag that indicates it is a record revoking state via lease
   3941  *     expiration or similar
   3942  *     OR has successfully acquired state.
   3943  * These structures vary in length, with the client string at the end, up
   3944  * to NFSV4_OPAQUELIMIT in size.
   3945  *
   3946  * At the end of the grace period, the file is truncated, the first
   3947  * record is rewritten with updated information and any acquired state
   3948  * records for successful reclaims of state are written.
   3949  *
   3950  * Subsequent records are appended when the first state is issued to
   3951  * a client and when state is revoked for a client.
   3952  *
   3953  * When reading the file in, state issued records that come later in
   3954  * the file override older ones, since the append log is in cronological order.
   3955  * If, for some reason, the file can't be read, the grace period is
   3956  * immediately terminated and all reclaims get NFSERR_NOGRACE.
   3957  */
   3958 
   3959 /*
   3960  * Read in the stable storage file. Called by nfssvc() before the nfsd
   3961  * processes start servicing requests.
   3962  */
   3963 APPLESTATIC void
   3964 nfsrv_setupstable(NFSPROC_T *p)
   3965 {
   3966 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
   3967 	struct nfsrv_stable *sp, *nsp;
   3968 	struct nfst_rec *tsp;
   3969 	int error, i, tryagain;
   3970 	off_t off = 0;
   3971 	ssize_t aresid, len;
   3972 
   3973 	/*
   3974 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
   3975 	 * a reboot, so state has not been lost.
   3976 	 */
   3977 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
   3978 		return;
   3979 	/*
   3980 	 * Set Grace over just until the file reads successfully.
   3981 	 */
   3982 	nfsrvboottime = time_second;
   3983 	LIST_INIT(&sf->nsf_head);
   3984 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
   3985 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
   3986 	if (sf->nsf_fp == NULL)
   3987 		return;
   3988 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
   3989 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
   3990 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
   3991 	if (error || aresid || sf->nsf_numboots == 0 ||
   3992 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
   3993 		return;
   3994 
   3995 	/*
   3996 	 * Now, read in the boottimes.
   3997 	 */
   3998 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
   3999 		sizeof (time_t), M_TEMP, M_WAITOK);
   4000 	off = sizeof (struct nfsf_rec);
   4001 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
   4002 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
   4003 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
   4004 	if (error || aresid) {
   4005 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
   4006 		sf->nsf_bootvals = NULL;
   4007 		return;
   4008 	}
   4009 
   4010 	/*
   4011 	 * Make sure this nfsrvboottime is different from all recorded
   4012 	 * previous ones.
   4013 	 */
   4014 	do {
   4015 		tryagain = 0;
   4016 		for (i = 0; i < sf->nsf_numboots; i++) {
   4017 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
   4018 				nfsrvboottime++;
   4019 				tryagain = 1;
   4020 				break;
   4021 			}
   4022 		}
   4023 	} while (tryagain);
   4024 
   4025 	sf->nsf_flags |= NFSNSF_OK;
   4026 	off += (sf->nsf_numboots * sizeof (time_t));
   4027 
   4028 	/*
   4029 	 * Read through the file, building a list of records for grace
   4030 	 * checking.
   4031 	 * Each record is between sizeof (struct nfst_rec) and
   4032 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
   4033 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
   4034 	 */
   4035 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
   4036 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
   4037 	do {
   4038 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
   4039 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
   4040 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
   4041 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
   4042 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
   4043 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
   4044 		/*
   4045 		 * Yuck, the file has been corrupted, so just return
   4046 		 * after clearing out any restart state, so the grace period
   4047 		 * is over.
   4048 		 */
   4049 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
   4050 			LIST_REMOVE(sp, nst_list);
   4051 			free((caddr_t)sp, M_TEMP);
   4052 		}
   4053 		free((caddr_t)tsp, M_TEMP);
   4054 		sf->nsf_flags &= ~NFSNSF_OK;
   4055 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
   4056 		sf->nsf_bootvals = NULL;
   4057 		return;
   4058 	    }
   4059 	    if (len > 0) {
   4060 		off += sizeof (struct nfst_rec) + tsp->len - 1;
   4061 		/*
   4062 		 * Search the list for a matching client.
   4063 		 */
   4064 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
   4065 			if (tsp->len == sp->nst_len &&
   4066 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
   4067 				break;
   4068 		}
   4069 		if (sp == LIST_END(&sf->nsf_head)) {
   4070 			sp = (struct nfsrv_stable *)malloc(tsp->len +
   4071 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
   4072 				M_WAITOK);
   4073 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
   4074 				sizeof (struct nfst_rec) + tsp->len - 1);
   4075 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
   4076 		} else {
   4077 			if (tsp->flag == NFSNST_REVOKE)
   4078 				sp->nst_flag |= NFSNST_REVOKE;
   4079 			else
   4080 				/*
   4081 				 * A subsequent timestamp indicates the client
   4082 				 * did a setclientid/confirm and any previous
   4083 				 * revoke is no longer relevant.
   4084 				 */
   4085 				sp->nst_flag &= ~NFSNST_REVOKE;
   4086 		}
   4087 	    }
   4088 	} while (len > 0);
   4089 	free((caddr_t)tsp, M_TEMP);
   4090 	sf->nsf_flags = NFSNSF_OK;
   4091 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
   4092 		NFSRV_LEASEDELTA;
   4093 }
   4094 
   4095 /*
   4096  * Update the stable storage file, now that the grace period is over.
   4097  */
   4098 APPLESTATIC void
   4099 nfsrv_updatestable(NFSPROC_T *p)
   4100 {
   4101 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
   4102 	struct nfsrv_stable *sp, *nsp;
   4103 	int i;
   4104 	struct nfsvattr nva;
   4105 	vnode_t vp;
   4106 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
   4107 	mount_t mp = NULL;
   4108 #endif
   4109 	int error;
   4110 
   4111 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
   4112 		return;
   4113 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
   4114 	/*
   4115 	 * Ok, we need to rewrite the stable storage file.
   4116 	 * - truncate to 0 length
   4117 	 * - write the new first structure
   4118 	 * - loop through the data structures, writing out any that
   4119 	 *   have timestamps older than the old boot
   4120 	 */
   4121 	if (sf->nsf_bootvals) {
   4122 		sf->nsf_numboots++;
   4123 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
   4124 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
   4125 	} else {
   4126 		sf->nsf_numboots = 1;
   4127 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
   4128 			M_TEMP, M_WAITOK);
   4129 	}
   4130 	sf->nsf_bootvals[0] = nfsrvboottime;
   4131 	sf->nsf_lease = nfsrv_lease;
   4132 	NFSVNO_ATTRINIT(&nva);
   4133 	NFSVNO_SETATTRVAL(&nva, size, 0);
   4134 	vp = NFSFPVNODE(sf->nsf_fp);
   4135 	vn_start_write(vp, &mp, V_WAIT);
   4136 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
   4137 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
   4138 		    NULL);
   4139 		NFSVOPUNLOCK(vp, 0);
   4140 	} else
   4141 		error = EPERM;
   4142 	vn_finished_write(mp);
   4143 	if (!error)
   4144 	    error = NFSD_RDWR(UIO_WRITE, vp,
   4145 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
   4146 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
   4147 	if (!error)
   4148 	    error = NFSD_RDWR(UIO_WRITE, vp,
   4149 		(caddr_t)sf->nsf_bootvals,
   4150 		sf->nsf_numboots * sizeof (time_t),
   4151 		(off_t)(sizeof (struct nfsf_rec)),
   4152 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
   4153 	free((caddr_t)sf->nsf_bootvals, M_TEMP);
   4154 	sf->nsf_bootvals = NULL;
   4155 	if (error) {
   4156 		sf->nsf_flags &= ~NFSNSF_OK;
   4157 		printf("EEK! Can't write NfsV4 stable storage file\n");
   4158 		return;
   4159 	}
   4160 	sf->nsf_flags |= NFSNSF_OK;
   4161 
   4162 	/*
   4163 	 * Loop through the list and write out timestamp records for
   4164 	 * any clients that successfully reclaimed state.
   4165 	 */
   4166 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
   4167 		if (sp->nst_flag & NFSNST_GOTSTATE) {
   4168 			nfsrv_writestable(sp->nst_client, sp->nst_len,
   4169 				NFSNST_NEWSTATE, p);
   4170 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
   4171 		}
   4172 		LIST_REMOVE(sp, nst_list);
   4173 		free((caddr_t)sp, M_TEMP);
   4174 	}
   4175 	nfsrv_backupstable();
   4176 }
   4177 
   4178 /*
   4179  * Append a record to the stable storage file.
   4180  */
   4181 APPLESTATIC void
   4182 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
   4183 {
   4184 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
   4185 	struct nfst_rec *sp;
   4186 	int error;
   4187 
   4188 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
   4189 		return;
   4190 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
   4191 		len - 1, M_TEMP, M_WAITOK);
   4192 	sp->len = len;
   4193 	NFSBCOPY(client, sp->client, len);
   4194 	sp->flag = flag;
   4195 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
   4196 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
   4197 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
   4198 	free((caddr_t)sp, M_TEMP);
   4199 	if (error) {
   4200 		sf->nsf_flags &= ~NFSNSF_OK;
   4201 		printf("EEK! Can't write NfsV4 stable storage file\n");
   4202 	}
   4203 }
   4204 
   4205 /*
   4206  * This function is called during the grace period to mark a client
   4207  * that successfully reclaimed state.
   4208  */
   4209 static void
   4210 nfsrv_markstable(struct nfsclient *clp)
   4211 {
   4212 	struct nfsrv_stable *sp;
   4213 
   4214 	/*
   4215 	 * First find the client structure.
   4216 	 */
   4217 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
   4218 		if (sp->nst_len == clp->lc_idlen &&
   4219 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
   4220 			break;
   4221 	}
   4222 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
   4223 		return;
   4224 
   4225 	/*
   4226 	 * Now, just mark it and set the nfsclient back pointer.
   4227 	 */
   4228 	sp->nst_flag |= NFSNST_GOTSTATE;
   4229 	sp->nst_clp = clp;
   4230 }
   4231 
   4232 /*
   4233  * This function is called for a reclaim, to see if it gets grace.
   4234  * It returns 0 if a reclaim is allowed, 1 otherwise.
   4235  */
   4236 static int
   4237 nfsrv_checkstable(struct nfsclient *clp)
   4238 {
   4239 	struct nfsrv_stable *sp;
   4240 
   4241 	/*
   4242 	 * First, find the entry for the client.
   4243 	 */
   4244 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
   4245 		if (sp->nst_len == clp->lc_idlen &&
   4246 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
   4247 			break;
   4248 	}
   4249 
   4250 	/*
   4251 	 * If not in the list, state was revoked or no state was issued
   4252 	 * since the previous reboot, a reclaim is denied.
   4253 	 */
   4254 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
   4255 	    (sp->nst_flag & NFSNST_REVOKE) ||
   4256 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
   4257 		return (1);
   4258 	return (0);
   4259 }
   4260 
   4261 /*
   4262  * Test for and try to clear out a conflicting client. This is called by
   4263  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
   4264  * a found.
   4265  * The trick here is that it can't revoke a conflicting client with an
   4266  * expired lease unless it holds the v4root lock, so...
   4267  * If no v4root lock, get the lock and return 1 to indicate "try again".
   4268  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
   4269  * the revocation worked and the conflicting client is "bye, bye", so it
   4270  * can be tried again.
   4271  * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
   4272  * Unlocks State before a non-zero value is returned.
   4273  */
   4274 static int
   4275 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
   4276     NFSPROC_T *p)
   4277 {
   4278 	int gotlock, lktype;
   4279 
   4280 	/*
   4281 	 * If lease hasn't expired, we can't fix it.
   4282 	 */
   4283 	if (clp->lc_expiry >= NFSD_MONOSEC ||
   4284 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
   4285 		return (0);
   4286 	if (*haslockp == 0) {
   4287 		NFSUNLOCKSTATE();
   4288 		lktype = NFSVOPISLOCKED(vp);
   4289 		NFSVOPUNLOCK(vp, 0);
   4290 		NFSLOCKV4ROOTMUTEX();
   4291 		nfsv4_relref(&nfsv4rootfs_lock);
   4292 		do {
   4293 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
   4294 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
   4295 		} while (!gotlock);
   4296 		NFSUNLOCKV4ROOTMUTEX();
   4297 		*haslockp = 1;
   4298 		NFSVOPLOCK(vp, lktype | LK_RETRY);
   4299 		if ((vp->v_iflag & VI_DOOMED) != 0)
   4300 			return (2);
   4301 		else
   4302 			return (1);
   4303 	}
   4304 	NFSUNLOCKSTATE();
   4305 
   4306 	/*
   4307 	 * Ok, we can expire the conflicting client.
   4308 	 */
   4309 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
   4310 	nfsrv_backupstable();
   4311 	nfsrv_cleanclient(clp, p);
   4312 	nfsrv_freedeleglist(&clp->lc_deleg);
   4313 	nfsrv_freedeleglist(&clp->lc_olddeleg);
   4314 	LIST_REMOVE(clp, lc_hash);
   4315 	nfsrv_zapclient(clp, p);
   4316 	return (1);
   4317 }
   4318 
   4319 /*
   4320  * Resolve a delegation conflict.
   4321  * Returns 0 to indicate the conflict was resolved without sleeping.
   4322  * Return -1 to indicate that the caller should check for conflicts again.
   4323  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
   4324  *
   4325  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
   4326  * for a return of 0, since there was no sleep and it could be required
   4327  * later. It is released for a return of NFSERR_DELAY, since the caller
   4328  * will return that error. It is released when a sleep was done waiting
   4329  * for the delegation to be returned or expire (so that other nfsds can
   4330  * handle ops). Then, it must be acquired for the write to stable storage.
   4331  * (This function is somewhat similar to nfsrv_clientconflict(), but
   4332  *  the semantics differ in a couple of subtle ways. The return of 0
   4333  *  indicates the conflict was resolved without sleeping here, not
   4334  *  that the conflict can't be resolved and the handling of nfsv4root_lock
   4335  *  differs, as noted above.)
   4336  * Unlocks State before returning a non-zero value.
   4337  */
   4338 static int
   4339 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
   4340     vnode_t vp)
   4341 {
   4342 	struct nfsclient *clp = stp->ls_clp;
   4343 	int gotlock, error, lktype, retrycnt, zapped_clp;
   4344 	nfsv4stateid_t tstateid;
   4345 	fhandle_t tfh;
   4346 
   4347 	/*
   4348 	 * If the conflict is with an old delegation...
   4349 	 */
   4350 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
   4351 		/*
   4352 		 * You can delete it, if it has expired.
   4353 		 */
   4354 		if (clp->lc_delegtime < NFSD_MONOSEC) {
   4355 			nfsrv_freedeleg(stp);
   4356 			NFSUNLOCKSTATE();
   4357 			error = -1;
   4358 			goto out;
   4359 		}
   4360 		NFSUNLOCKSTATE();
   4361 		/*
   4362 		 * During this delay, the old delegation could expire or it
   4363 		 * could be recovered by the client via an Open with
   4364 		 * CLAIM_DELEGATE_PREV.
   4365 		 * Release the nfsv4root_lock, if held.
   4366 		 */
   4367 		if (*haslockp) {
   4368 			*haslockp = 0;
   4369 			NFSLOCKV4ROOTMUTEX();
   4370 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
   4371 			NFSUNLOCKV4ROOTMUTEX();
   4372 		}
   4373 		error = NFSERR_DELAY;
   4374 		goto out;
   4375 	}
   4376 
   4377 	/*
   4378 	 * It's a current delegation, so:
   4379 	 * - check to see if the delegation has expired
   4380 	 *   - if so, get the v4root lock and then expire it
   4381 	 */
   4382 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
   4383 		/*
   4384 		 * - do a recall callback, since not yet done
   4385 		 * For now, never allow truncate to be set. To use
   4386 		 * truncate safely, it must be guaranteed that the
   4387 		 * Remove, Rename or Setattr with size of 0 will
   4388 		 * succeed and that would require major changes to
   4389 		 * the VFS/Vnode OPs.
   4390 		 * Set the expiry time large enough so that it won't expire
   4391 		 * until after the callback, then set it correctly, once
   4392 		 * the callback is done. (The delegation will now time
   4393 		 * out whether or not the Recall worked ok. The timeout
   4394 		 * will be extended when ops are done on the delegation
   4395 		 * stateid, up to the timelimit.)
   4396 		 */
   4397 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
   4398 		    NFSRV_LEASEDELTA;
   4399 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
   4400 		    NFSRV_LEASEDELTA;
   4401 		stp->ls_flags |= NFSLCK_DELEGRECALL;
   4402 
   4403 		/*
   4404 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
   4405 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
   4406 		 * in order to try and avoid a race that could happen
   4407 		 * when a CBRecall request passed the Open reply with
   4408 		 * the delegation in it when transitting the network.
   4409 		 * Since nfsrv_docallback will sleep, don't use stp after
   4410 		 * the call.
   4411 		 */
   4412 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
   4413 		    sizeof (tstateid));
   4414 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
   4415 		    sizeof (tfh));
   4416 		NFSUNLOCKSTATE();
   4417 		if (*haslockp) {
   4418 			*haslockp = 0;
   4419 			NFSLOCKV4ROOTMUTEX();
   4420 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
   4421 			NFSUNLOCKV4ROOTMUTEX();
   4422 		}
   4423 		retrycnt = 0;
   4424 		do {
   4425 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
   4426 			&tstateid, 0, &tfh, NULL, NULL, p);
   4427 		    retrycnt++;
   4428 		} while ((error == NFSERR_BADSTATEID ||
   4429 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
   4430 		error = NFSERR_DELAY;
   4431 		goto out;
   4432 	}
   4433 
   4434 	if (clp->lc_expiry >= NFSD_MONOSEC &&
   4435 	    stp->ls_delegtime >= NFSD_MONOSEC) {
   4436 		NFSUNLOCKSTATE();
   4437 		/*
   4438 		 * A recall has been done, but it has not yet expired.
   4439 		 * So, RETURN_DELAY.
   4440 		 */
   4441 		if (*haslockp) {
   4442 			*haslockp = 0;
   4443 			NFSLOCKV4ROOTMUTEX();
   4444 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
   4445 			NFSUNLOCKV4ROOTMUTEX();
   4446 		}
   4447 		error = NFSERR_DELAY;
   4448 		goto out;
   4449 	}
   4450 
   4451 	/*
   4452 	 * If we don't yet have the lock, just get it and then return,
   4453 	 * since we need that before deleting expired state, such as
   4454 	 * this delegation.
   4455 	 * When getting the lock, unlock the vnode, so other nfsds that
   4456 	 * are in progress, won't get stuck waiting for the vnode lock.
   4457 	 */
   4458 	if (*haslockp == 0) {
   4459 		NFSUNLOCKSTATE();
   4460 		lktype = NFSVOPISLOCKED(vp);
   4461 		NFSVOPUNLOCK(vp, 0);
   4462 		NFSLOCKV4ROOTMUTEX();
   4463 		nfsv4_relref(&nfsv4rootfs_lock);
   4464 		do {
   4465 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
   4466 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
   4467 		} while (!gotlock);
   4468 		NFSUNLOCKV4ROOTMUTEX();
   4469 		*haslockp = 1;
   4470 		NFSVOPLOCK(vp, lktype | LK_RETRY);
   4471 		if ((vp->v_iflag & VI_DOOMED) != 0) {
   4472 			*haslockp = 0;
   4473 			NFSLOCKV4ROOTMUTEX();
   4474 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
   4475 			NFSUNLOCKV4ROOTMUTEX();
   4476 			error = NFSERR_PERM;
   4477 			goto out;
   4478 		}
   4479 		error = -1;
   4480 		goto out;
   4481 	}
   4482 
   4483 	NFSUNLOCKSTATE();
   4484 	/*
   4485 	 * Ok, we can delete the expired delegation.
   4486 	 * First, write the Revoke record to stable storage and then
   4487 	 * clear out the conflict.
   4488 	 * Since all other nfsd threads are now blocked, we can safely
   4489 	 * sleep without the state changing.
   4490 	 */
   4491 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
   4492 	nfsrv_backupstable();
   4493 	if (clp->lc_expiry < NFSD_MONOSEC) {
   4494 		nfsrv_cleanclient(clp, p);
   4495 		nfsrv_freedeleglist(&clp->lc_deleg);
   4496 		nfsrv_freedeleglist(&clp->lc_olddeleg);
   4497 		LIST_REMOVE(clp, lc_hash);
   4498 		zapped_clp = 1;
   4499 	} else {
   4500 		nfsrv_freedeleg(stp);
   4501 		zapped_clp = 0;
   4502 	}
   4503 	if (zapped_clp)
   4504 		nfsrv_zapclient(clp, p);
   4505 	error = -1;
   4506 
   4507 out:
   4508 	NFSEXITCODE(error);
   4509 	return (error);
   4510 }
   4511 
   4512 /*
   4513  * Check for a remove allowed, if remove is set to 1 and get rid of
   4514  * delegations.
   4515  */
   4516 APPLESTATIC int
   4517 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
   4518 {
   4519 	struct nfsstate *stp;
   4520 	struct nfslockfile *lfp;
   4521 	int error, haslock = 0;
   4522 	fhandle_t nfh;
   4523 
   4524 	/*
   4525 	 * First, get the lock file structure.
   4526 	 * (A return of -1 means no associated state, so remove ok.)
   4527 	 */
   4528 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
   4529 tryagain:
   4530 	NFSLOCKSTATE();
   4531 	if (!error)
   4532 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
   4533 	if (error) {
   4534 		NFSUNLOCKSTATE();
   4535 		if (haslock) {
   4536 			NFSLOCKV4ROOTMUTEX();
   4537 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
   4538 			NFSUNLOCKV4ROOTMUTEX();
   4539 		}
   4540 		if (error == -1)
   4541 			error = 0;
   4542 		goto out;
   4543 	}
   4544 
   4545 	/*
   4546 	 * Now, we must Recall any delegations.
   4547 	 */
   4548 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
   4549 	if (error) {
   4550 		/*
   4551 		 * nfsrv_cleandeleg() unlocks state for non-zero
   4552 		 * return.
   4553 		 */
   4554 		if (error == -1)
   4555 			goto tryagain;
   4556 		if (haslock) {
   4557 			NFSLOCKV4ROOTMUTEX();
   4558 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
   4559 			NFSUNLOCKV4ROOTMUTEX();
   4560 		}
   4561 		goto out;
   4562 	}
   4563 
   4564 	/*
   4565 	 * Now, look for a conflicting open share.
   4566 	 */
   4567 	if (remove) {
   4568 		LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
   4569 			if (stp->ls_flags & NFSLCK_WRITEDENY) {
   4570 				error = NFSERR_FILEOPEN;
   4571 				break;
   4572 			}
   4573 		}
   4574 	}
   4575 
   4576 	NFSUNLOCKSTATE();
   4577 	if (haslock) {
   4578 		NFSLOCKV4ROOTMUTEX();
   4579 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
   4580 		NFSUNLOCKV4ROOTMUTEX();
   4581 	}
   4582 
   4583 out:
   4584 	NFSEXITCODE(error);
   4585 	return (error);
   4586 }
   4587 
   4588 /*
   4589  * Clear out all delegations for the file referred to by lfp.
   4590  * May return NFSERR_DELAY, if there will be a delay waiting for
   4591  * delegations to expire.
   4592  * Returns -1 to indicate it slept while recalling a delegation.
   4593  * This function has the side effect of deleting the nfslockfile structure,
   4594  * if it no longer has associated state and didn't have to sleep.
   4595  * Unlocks State before a non-zero value is returned.
   4596  */
   4597 static int
   4598 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
   4599     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
   4600 {
   4601 	struct nfsstate *stp, *nstp;
   4602 	int ret = 0;
   4603 
   4604 	stp = LIST_FIRST(&lfp->lf_deleg);
   4605 	while (stp != LIST_END(&lfp->lf_deleg)) {
   4606 		nstp = LIST_NEXT(stp, ls_file);
   4607 		if (stp->ls_clp != clp) {
   4608 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
   4609 			if (ret) {
   4610 				/*
   4611 				 * nfsrv_delegconflict() unlocks state
   4612 				 * when it returns non-zero.
   4613 				 */
   4614 				goto out;
   4615 			}
   4616 		}
   4617 		stp = nstp;
   4618 	}
   4619 out:
   4620 	NFSEXITCODE(ret);
   4621 	return (ret);
   4622 }
   4623 
   4624 /*
   4625  * There are certain operations that, when being done outside of NFSv4,
   4626  * require that any NFSv4 delegation for the file be recalled.
   4627  * This function is to be called for those cases:
   4628  * VOP_RENAME() - When a delegation is being recalled for any reason,
   4629  *	the client may have to do Opens against the server, using the file's
   4630  *	final component name. If the file has been renamed on the server,
   4631  *	that component name will be incorrect and the Open will fail.
   4632  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
   4633  *	been removed on the server, if there is a delegation issued to
   4634  *	that client for the file. I say "theoretically" since clients
   4635  *	normally do an Access Op before the Open and that Access Op will
   4636  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
   4637  *	they will detect the file's removal in the same manner. (There is
   4638  *	one case where RFC3530 allows a client to do an Open without first
   4639  *	doing an Access Op, which is passage of a check against the ACE
   4640  *	returned with a Write delegation, but current practice is to ignore
   4641  *	the ACE and always do an Access Op.)
   4642  *	Since the functions can only be called with an unlocked vnode, this
   4643  *	can't be done at this time.
   4644  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
   4645  *	locks locally in the client, which are not visible to the server. To
   4646  *	deal with this, issuing of delegations for a vnode must be disabled
   4647  *	and all delegations for the vnode recalled. This is done via the
   4648  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
   4649  */
   4650 APPLESTATIC void
   4651 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
   4652 {
   4653 	time_t starttime;
   4654 	int error;
   4655 
   4656 	/*
   4657 	 * First, check to see if the server is currently running and it has
   4658 	 * been called for a regular file when issuing delegations.
   4659 	 */
   4660 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
   4661 	    nfsrv_issuedelegs == 0)
   4662 		return;
   4663 
   4664 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
   4665 	/*
   4666 	 * First, get a reference on the nfsv4rootfs_lock so that an
   4667 	 * exclusive lock cannot be acquired by another thread.
   4668 	 */
   4669 	NFSLOCKV4ROOTMUTEX();
   4670 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
   4671 	NFSUNLOCKV4ROOTMUTEX();
   4672 
   4673 	/*
   4674 	 * Now, call nfsrv_checkremove() in a loop while it returns
   4675 	 * NFSERR_DELAY. Return upon any other error or when timed out.
   4676 	 */
   4677 	starttime = NFSD_MONOSEC;
   4678 	do {
   4679 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
   4680 			error = nfsrv_checkremove(vp, 0, p);
   4681 			NFSVOPUNLOCK(vp, 0);
   4682 		} else
   4683 			error = EPERM;
   4684 		if (error == NFSERR_DELAY) {
   4685 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
   4686 				break;
   4687 			/* Sleep for a short period of time */
   4688 			(void) nfs_catnap(PZERO, 0, "nfsremove");
   4689 		}
   4690 	} while (error == NFSERR_DELAY);
   4691 	NFSLOCKV4ROOTMUTEX();
   4692 	nfsv4_relref(&nfsv4rootfs_lock);
   4693 	NFSUNLOCKV4ROOTMUTEX();
   4694 }
   4695 
   4696 APPLESTATIC void
   4697 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
   4698 {
   4699 
   4700 #ifdef VV_DISABLEDELEG
   4701 	/*
   4702 	 * First, flag issuance of delegations disabled.
   4703 	 */
   4704 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
   4705 #endif
   4706 
   4707 	/*
   4708 	 * Then call nfsd_recalldelegation() to get rid of all extant
   4709 	 * delegations.
   4710 	 */
   4711 	nfsd_recalldelegation(vp, p);
   4712 }
   4713 
   4714 /*
   4715  * Check for conflicting locks, etc. and then get rid of delegations.
   4716  * (At one point I thought that I should get rid of delegations for any
   4717  *  Setattr, since it could potentially disallow the I/O op (read or write)
   4718  *  allowed by the delegation. However, Setattr Ops that aren't changing
   4719  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
   4720  *  for the same client or a different one, so I decided to only get rid
   4721  *  of delegations for other clients when the size is being changed.)
   4722  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
   4723  * as Write backs, even if there is no delegation, so it really isn't any
   4724  * different?)
   4725  */
   4726 APPLESTATIC int
   4727 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
   4728     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
   4729     struct nfsexstuff *exp, NFSPROC_T *p)
   4730 {
   4731 	struct nfsstate st, *stp = &st;
   4732 	struct nfslock lo, *lop = &lo;
   4733 	int error = 0;
   4734 	nfsquad_t clientid;
   4735 
   4736 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
   4737 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
   4738 		lop->lo_first = nvap->na_size;
   4739 	} else {
   4740 		stp->ls_flags = 0;
   4741 		lop->lo_first = 0;
   4742 	}
   4743 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
   4744 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
   4745 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
   4746 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
   4747 		stp->ls_flags |= NFSLCK_SETATTR;
   4748 	if (stp->ls_flags == 0)
   4749 		goto out;
   4750 	lop->lo_end = NFS64BITSSET;
   4751 	lop->lo_flags = NFSLCK_WRITE;
   4752 	stp->ls_ownerlen = 0;
   4753 	stp->ls_op = NULL;
   4754 	stp->ls_uid = nd->nd_cred->cr_uid;
   4755 	stp->ls_stateid.seqid = stateidp->seqid;
   4756 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
   4757 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
   4758 	stp->ls_stateid.other[2] = stateidp->other[2];
   4759 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
   4760 	    stateidp, exp, nd, p);
   4761 
   4762 out:
   4763 	NFSEXITCODE2(error, nd);
   4764 	return (error);
   4765 }
   4766 
   4767 /*
   4768  * Check for a write delegation and do a CBGETATTR if there is one, updating
   4769  * the attributes, as required.
   4770  * Should I return an error if I can't get the attributes? (For now, I'll
   4771  * just return ok.
   4772  */
   4773 APPLESTATIC int
   4774 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
   4775     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
   4776     NFSPROC_T *p)
   4777 {
   4778 	struct nfsstate *stp;
   4779 	struct nfslockfile *lfp;
   4780 	struct nfsclient *clp;
   4781 	struct nfsvattr nva;
   4782 	fhandle_t nfh;
   4783 	int error = 0;
   4784 	nfsattrbit_t cbbits;
   4785 	u_quad_t delegfilerev;
   4786 
   4787 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
   4788 	if (!NFSNONZERO_ATTRBIT(&cbbits))
   4789 		goto out;
   4790 
   4791 	/*
   4792 	 * Get the lock file structure.
   4793 	 * (A return of -1 means no associated state, so return ok.)
   4794 	 */
   4795 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
   4796 	NFSLOCKSTATE();
   4797 	if (!error)
   4798 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
   4799 	if (error) {
   4800 		NFSUNLOCKSTATE();
   4801 		if (error == -1)
   4802 			error = 0;
   4803 		goto out;
   4804 	}
   4805 
   4806 	/*
   4807 	 * Now, look for a write delegation.
   4808 	 */
   4809 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
   4810 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
   4811 			break;
   4812 	}
   4813 	if (stp == LIST_END(&lfp->lf_deleg)) {
   4814 		NFSUNLOCKSTATE();
   4815 		goto out;
   4816 	}
   4817 	clp = stp->ls_clp;
   4818 	delegfilerev = stp->ls_filerev;
   4819 
   4820 	/*
   4821 	 * If the Write delegation was issued as a part of this Compound RPC
   4822 	 * or if we have an Implied Clientid (used in a previous Op in this
   4823 	 * compound) and it is the client the delegation was issued to,
   4824 	 * just return ok.
   4825 	 * I also assume that it is from the same client iff the network
   4826 	 * host IP address is the same as the callback address. (Not
   4827 	 * exactly correct by the RFC, but avoids a lot of Getattr
   4828 	 * callbacks.)
   4829 	 */
   4830 	if (nd->nd_compref == stp->ls_compref ||
   4831 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
   4832 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
   4833 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
   4834 		NFSUNLOCKSTATE();
   4835 		goto out;
   4836 	}
   4837 
   4838 	/*
   4839 	 * We are now done with the delegation state structure,
   4840 	 * so the statelock can be released and we can now tsleep().
   4841 	 */
   4842 
   4843 	/*
   4844 	 * Now, we must do the CB Getattr callback, to see if Change or Size
   4845 	 * has changed.
   4846 	 */
   4847 	if (clp->lc_expiry >= NFSD_MONOSEC) {
   4848 		NFSUNLOCKSTATE();
   4849 		NFSVNO_ATTRINIT(&nva);
   4850 		nva.na_filerev = NFS64BITSSET;
   4851 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
   4852 		    0, &nfh, &nva, &cbbits, p);
   4853 		if (!error) {
   4854 			if ((nva.na_filerev != NFS64BITSSET &&
   4855 			    nva.na_filerev > delegfilerev) ||
   4856 			    (NFSVNO_ISSETSIZE(&nva) &&
   4857 			     nva.na_size != nvap->na_size)) {
   4858 				nfsvno_updfilerev(vp, nvap, cred, p);
   4859 				if (NFSVNO_ISSETSIZE(&nva))
   4860 					nvap->na_size = nva.na_size;
   4861 			}
   4862 		}
   4863 	} else {
   4864 		NFSUNLOCKSTATE();
   4865 	}
   4866 	error = 0;
   4867 
   4868 out:
   4869 	NFSEXITCODE2(error, nd);
   4870 	return (error);
   4871 }
   4872 
   4873 /*
   4874  * This function looks for openowners that haven't had any opens for
   4875  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
   4876  * is set.
   4877  */
   4878 APPLESTATIC void
   4879 nfsrv_throwawayopens(NFSPROC_T *p)
   4880 {
   4881 	struct nfsclient *clp, *nclp;
   4882 	struct nfsstate *stp, *nstp;
   4883 	int i;
   4884 
   4885 	NFSLOCKSTATE();
   4886 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
   4887 	/*
   4888 	 * For each client...
   4889 	 */
   4890 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
   4891 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
   4892 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
   4893 			if (LIST_EMPTY(&stp->ls_open) &&
   4894 			    (stp->ls_noopens > NFSNOOPEN ||
   4895 			     (nfsrv_openpluslock * 2) >
   4896 			     NFSRV_V4STATELIMIT))
   4897 				nfsrv_freeopenowner(stp, 0, p);
   4898 		}
   4899 	    }
   4900 	}
   4901 	NFSUNLOCKSTATE();
   4902 }
   4903 
   4904 /*
   4905  * This function checks to see if the credentials are the same.
   4906  * Returns 1 for not same, 0 otherwise.
   4907  */
   4908 static int
   4909 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
   4910 {
   4911 
   4912 	if (nd->nd_flag & ND_GSS) {
   4913 		if (!(clp->lc_flags & LCL_GSS))
   4914 			return (1);
   4915 		if (clp->lc_flags & LCL_NAME) {
   4916 			if (nd->nd_princlen != clp->lc_namelen ||
   4917 			    NFSBCMP(nd->nd_principal, clp->lc_name,
   4918 				clp->lc_namelen))
   4919 				return (1);
   4920 			else
   4921 				return (0);
   4922 		}
   4923 		if (nd->nd_cred->cr_uid == clp->lc_uid)
   4924 			return (0);
   4925 		else
   4926 			return (1);
   4927 	} else if (clp->lc_flags & LCL_GSS)
   4928 		return (1);
   4929 	/*
   4930 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
   4931 	 * in RFC3530, which talks about principals, but doesn't say anything
   4932 	 * about uids for AUTH_SYS.)
   4933 	 */
   4934 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
   4935 		return (0);
   4936 	else
   4937 		return (1);
   4938 }
   4939 
   4940 /*
   4941  * Calculate the lease expiry time.
   4942  */
   4943 static time_t
   4944 nfsrv_leaseexpiry(void)
   4945 {
   4946 
   4947 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
   4948 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
   4949 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
   4950 }
   4951 
   4952 /*
   4953  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
   4954  */
   4955 static void
   4956 nfsrv_delaydelegtimeout(struct nfsstate *stp)
   4957 {
   4958 
   4959 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
   4960 		return;
   4961 
   4962 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
   4963 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
   4964 		stp->ls_delegtime += nfsrv_lease;
   4965 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
   4966 			stp->ls_delegtime = stp->ls_delegtimelimit;
   4967 	}
   4968 }
   4969 
   4970 /*
   4971  * This function checks to see if there is any other state associated
   4972  * with the openowner for this Open.
   4973  * It returns 1 if there is no other state, 0 otherwise.
   4974  */
   4975 static int
   4976 nfsrv_nootherstate(struct nfsstate *stp)
   4977 {
   4978 	struct nfsstate *tstp;
   4979 
   4980 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
   4981 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
   4982 			return (0);
   4983 	}
   4984 	return (1);
   4985 }
   4986 
   4987 /*
   4988  * Create a list of lock deltas (changes to local byte range locking
   4989  * that can be rolled back using the list) and apply the changes via
   4990  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
   4991  * the rollback or update function will be called after this.
   4992  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
   4993  * call fails. If it returns an error, it will unlock the list.
   4994  */
   4995 static int
   4996 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
   4997     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
   4998 {
   4999 	struct nfslock *lop, *nlop;
   5000 	int error = 0;
   5001 
   5002 	/* Loop through the list of locks. */
   5003 	lop = LIST_FIRST(&lfp->lf_locallock);
   5004 	while (first < end && lop != NULL) {
   5005 		nlop = LIST_NEXT(lop, lo_lckowner);
   5006 		if (first >= lop->lo_end) {
   5007 			/* not there yet */
   5008 			lop = nlop;
   5009 		} else if (first < lop->lo_first) {
   5010 			/* new one starts before entry in list */
   5011 			if (end <= lop->lo_first) {
   5012 				/* no overlap between old and new */
   5013 				error = nfsrv_dolocal(vp, lfp, flags,
   5014 				    NFSLCK_UNLOCK, first, end, cfp, p);
   5015 				if (error != 0)
   5016 					break;
   5017 				first = end;
   5018 			} else {
   5019 				/* handle fragment overlapped with new one */
   5020 				error = nfsrv_dolocal(vp, lfp, flags,
   5021 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
   5022 				    p);
   5023 				if (error != 0)
   5024 					break;
   5025 				first = lop->lo_first;
   5026 			}
   5027 		} else {
   5028 			/* new one overlaps this entry in list */
   5029 			if (end <= lop->lo_end) {
   5030 				/* overlaps all of new one */
   5031 				error = nfsrv_dolocal(vp, lfp, flags,
   5032 				    lop->lo_flags, first, end, cfp, p);
   5033 				if (error != 0)
   5034 					break;
   5035 				first = end;
   5036 			} else {
   5037 				/* handle fragment overlapped with new one */
   5038 				error = nfsrv_dolocal(vp, lfp, flags,
   5039 				    lop->lo_flags, first, lop->lo_end, cfp, p);
   5040 				if (error != 0)
   5041 					break;
   5042 				first = lop->lo_end;
   5043 				lop = nlop;
   5044 			}
   5045 		}
   5046 	}
   5047 	if (first < end && error == 0)
   5048 		/* handle fragment past end of list */
   5049 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
   5050 		    end, cfp, p);
   5051 
   5052 	NFSEXITCODE(error);
   5053 	return (error);
   5054 }
   5055 
   5056 /*
   5057  * Local lock unlock. Unlock all byte ranges that are no longer locked
   5058  * by NFSv4. To do this, unlock any subranges of first-->end that
   5059  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
   5060  * list. This list has all locks for the file held by other
   5061  * <clientid, lockowner> tuples. The list is ordered by increasing
   5062  * lo_first value, but may have entries that overlap each other, for
   5063  * the case of read locks.
   5064  */
   5065 static void
   5066 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
   5067     uint64_t init_end, NFSPROC_T *p)
   5068 {
   5069 	struct nfslock *lop;
   5070 	uint64_t first, end, prevfirst;
   5071 
   5072 	first = init_first;
   5073 	end = init_end;
   5074 	while (first < init_end) {
   5075 		/* Loop through all nfs locks, adjusting first and end */
   5076 		prevfirst = 0;
   5077 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
   5078 			KASSERT(prevfirst <= lop->lo_first,
   5079 			    ("nfsv4 locks out of order"));
   5080 			KASSERT(lop->lo_first < lop->lo_end,
   5081 			    ("nfsv4 bogus lock"));
   5082 			prevfirst = lop->lo_first;
   5083 			if (first >= lop->lo_first &&
   5084 			    first < lop->lo_end)
   5085 				/*
   5086 				 * Overlaps with initial part, so trim
   5087 				 * off that initial part by moving first past
   5088 				 * it.
   5089 				 */
   5090 				first = lop->lo_end;
   5091 			else if (end > lop->lo_first &&
   5092 			    lop->lo_first > first) {
   5093 				/*
   5094 				 * This lock defines the end of the
   5095 				 * segment to unlock, so set end to the
   5096 				 * start of it and break out of the loop.
   5097 				 */
   5098 				end = lop->lo_first;
   5099 				break;
   5100 			}
   5101 			if (first >= end)
   5102 				/*
   5103 				 * There is no segment left to do, so
   5104 				 * break out of this loop and then exit
   5105 				 * the outer while() since first will be set
   5106 				 * to end, which must equal init_end here.
   5107 				 */
   5108 				break;
   5109 		}
   5110 		if (first < end) {
   5111 			/* Unlock this segment */
   5112 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
   5113 			    NFSLCK_READ, first, end, NULL, p);
   5114 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
   5115 			    first, end);
   5116 		}
   5117 		/*
   5118 		 * Now move past this segment and look for any further
   5119 		 * segment in the range, if there is one.
   5120 		 */
   5121 		first = end;
   5122 		end = init_end;
   5123 	}
   5124 }
   5125 
   5126 /*
   5127  * Do the local lock operation and update the rollback list, as required.
   5128  * Perform the rollback and return the error if nfsvno_advlock() fails.
   5129  */
   5130 static int
   5131 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
   5132     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
   5133 {
   5134 	struct nfsrollback *rlp;
   5135 	int error = 0, ltype, oldltype;
   5136 
   5137 	if (flags & NFSLCK_WRITE)
   5138 		ltype = F_WRLCK;
   5139 	else if (flags & NFSLCK_READ)
   5140 		ltype = F_RDLCK;
   5141 	else
   5142 		ltype = F_UNLCK;
   5143 	if (oldflags & NFSLCK_WRITE)
   5144 		oldltype = F_WRLCK;
   5145 	else if (oldflags & NFSLCK_READ)
   5146 		oldltype = F_RDLCK;
   5147 	else
   5148 		oldltype = F_UNLCK;
   5149 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
   5150 		/* nothing to do */
   5151 		goto out;
   5152 	error = nfsvno_advlock(vp, ltype, first, end, p);
   5153 	if (error != 0) {
   5154 		if (cfp != NULL) {
   5155 			cfp->cl_clientid.lval[0] = 0;
   5156 			cfp->cl_clientid.lval[1] = 0;
   5157 			cfp->cl_first = 0;
   5158 			cfp->cl_end = NFS64BITSSET;
   5159 			cfp->cl_flags = NFSLCK_WRITE;
   5160 			cfp->cl_ownerlen = 5;
   5161 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
   5162 		}
   5163 		nfsrv_locallock_rollback(vp, lfp, p);
   5164 	} else if (ltype != F_UNLCK) {
   5165 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
   5166 		    M_WAITOK);
   5167 		rlp->rlck_first = first;
   5168 		rlp->rlck_end = end;
   5169 		rlp->rlck_type = oldltype;
   5170 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
   5171 	}
   5172 
   5173 out:
   5174 	NFSEXITCODE(error);
   5175 	return (error);
   5176 }
   5177 
   5178 /*
   5179  * Roll back local lock changes and free up the rollback list.
   5180  */
   5181 static void
   5182 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
   5183 {
   5184 	struct nfsrollback *rlp, *nrlp;
   5185 
   5186 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
   5187 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
   5188 		    rlp->rlck_end, p);
   5189 		free(rlp, M_NFSDROLLBACK);
   5190 	}
   5191 	LIST_INIT(&lfp->lf_rollback);
   5192 }
   5193 
   5194 /*
   5195  * Update local lock list and delete rollback list (ie now committed to the
   5196  * local locks). Most of the work is done by the internal function.
   5197  */
   5198 static void
   5199 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
   5200     uint64_t end)
   5201 {
   5202 	struct nfsrollback *rlp, *nrlp;
   5203 	struct nfslock *new_lop, *other_lop;
   5204 
   5205 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
   5206 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
   5207 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
   5208 		    M_WAITOK);
   5209 	else
   5210 		other_lop = NULL;
   5211 	new_lop->lo_flags = flags;
   5212 	new_lop->lo_first = first;
   5213 	new_lop->lo_end = end;
   5214 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
   5215 	if (new_lop != NULL)
   5216 		free(new_lop, M_NFSDLOCK);
   5217 	if (other_lop != NULL)
   5218 		free(other_lop, M_NFSDLOCK);
   5219 
   5220 	/* and get rid of the rollback list */
   5221 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
   5222 		free(rlp, M_NFSDROLLBACK);
   5223 	LIST_INIT(&lfp->lf_rollback);
   5224 }
   5225 
   5226 /*
   5227  * Lock the struct nfslockfile for local lock updating.
   5228  */
   5229 static void
   5230 nfsrv_locklf(struct nfslockfile *lfp)
   5231 {
   5232 	int gotlock;
   5233 
   5234 	/* lf_usecount ensures *lfp won't be free'd */
   5235 	lfp->lf_usecount++;
   5236 	do {
   5237 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
   5238 		    NFSSTATEMUTEXPTR, NULL);
   5239 	} while (gotlock == 0);
   5240 	lfp->lf_usecount--;
   5241 }
   5242 
   5243 /*
   5244  * Unlock the struct nfslockfile after local lock updating.
   5245  */
   5246 static void
   5247 nfsrv_unlocklf(struct nfslockfile *lfp)
   5248 {
   5249 
   5250 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
   5251 }
   5252 
   5253 /*
   5254  * Clear out all state for the NFSv4 server.
   5255  * Must be called by a thread that can sleep when no nfsds are running.
   5256  */
   5257 void
   5258 nfsrv_throwawayallstate(NFSPROC_T *p)
   5259 {
   5260 	struct nfsclient *clp, *nclp;
   5261 	struct nfslockfile *lfp, *nlfp;
   5262 	int i;
   5263 
   5264 	/*
   5265 	 * For each client, clean out the state and then free the structure.
   5266 	 */
   5267 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
   5268 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
   5269 			nfsrv_cleanclient(clp, p);
   5270 			nfsrv_freedeleglist(&clp->lc_deleg);
   5271 			nfsrv_freedeleglist(&clp->lc_olddeleg);
   5272 			free(clp, M_NFSDCLIENT);
   5273 		}
   5274 	}
   5275 
   5276 	/*
   5277 	 * Also, free up any remaining lock file structures.
   5278 	 */
   5279 	for (i = 0; i < NFSLOCKHASHSIZE; i++) {
   5280 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
   5281 			printf("nfsd unload: fnd a lock file struct\n");
   5282 			nfsrv_freenfslockfile(lfp);
   5283 		}
   5284 	}
   5285 }
   5286 
   5287