xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision f81cdf24ba5436367377f7c8e8f51f6df2a75ca7)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2009 Rick Macklem, University of Guelph
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  */
29 
30 #include <sys/cdefs.h>
31 #include "opt_inet.h"
32 #include "opt_inet6.h"
33 #include <sys/extattr.h>
34 #include <fs/nfs/nfsport.h>
35 
36 int nfsrv_issuedelegs = 0;
37 int nfsrv_dolocallocks = 0;
38 struct nfsv4lock nfsv4rootfs_lock;
39 time_t nfsdev_time = 0;
40 int nfsrv_layouthashsize;
41 volatile int nfsrv_layoutcnt = 0;
42 
43 NFSD_VNET_DEFINE(struct nfsrv_stablefirst, nfsrv_stablefirst);
44 
45 NFSD_VNET_DECLARE(int, nfsrv_numnfsd);
46 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p);
47 
48 extern uint32_t nfs_srvmaxio;
49 extern int nfsrv_lease;
50 extern struct timeval nfsboottime;
51 extern u_int32_t newnfs_true, newnfs_false;
52 extern struct mtx nfsrv_dslock_mtx;
53 extern struct mtx nfsrv_recalllock_mtx;
54 extern struct mtx nfsrv_dontlistlock_mtx;
55 extern int nfsd_debuglevel;
56 extern u_int nfsrv_dsdirsize;
57 extern struct nfsdevicehead nfsrv_devidhead;
58 extern int nfsrv_doflexfile;
59 extern int nfsrv_maxpnfsmirror;
60 NFSV4ROOTLOCKMUTEX;
61 NFSSTATESPINLOCK;
62 extern struct nfsdontlisthead nfsrv_dontlisthead;
63 extern volatile int nfsrv_devidcnt;
64 extern struct nfslayouthead nfsrv_recalllisthead;
65 extern char *nfsrv_zeropnfsdat;
66 
67 SYSCTL_DECL(_vfs_nfsd);
68 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
69 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
70     &nfsrv_statehashsize, 0,
71     "Size of state hash table set via loader.conf");
72 
73 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
74 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
75     &nfsrv_clienthashsize, 0,
76     "Size of client hash table set via loader.conf");
77 
78 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
79 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
80     &nfsrv_lockhashsize, 0,
81     "Size of file handle hash table set via loader.conf");
82 
83 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
84 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
85     &nfsrv_sessionhashsize, 0,
86     "Size of session hash table set via loader.conf");
87 
88 int	nfsrv_layouthighwater = NFSLAYOUTHIGHWATER;
89 SYSCTL_INT(_vfs_nfsd, OID_AUTO, layouthighwater, CTLFLAG_RDTUN,
90     &nfsrv_layouthighwater, 0,
91     "High water mark for number of layouts set via loader.conf");
92 
93 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
94 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
95     &nfsrv_v4statelimit, 0,
96     "High water limit for NFSv4 opens+locks+delegations");
97 
98 static int	nfsrv_writedelegifpos = 0;
99 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
100     &nfsrv_writedelegifpos, 0,
101     "Issue a write delegation for read opens if possible");
102 
103 static int	nfsrv_allowreadforwriteopen = 1;
104 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
105     &nfsrv_allowreadforwriteopen, 0,
106     "Allow Reads to be done with Write Access StateIDs");
107 
108 int	nfsrv_pnfsatime = 0;
109 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW,
110     &nfsrv_pnfsatime, 0,
111     "For pNFS service, do Getattr ops to keep atime up-to-date");
112 
113 int	nfsrv_flexlinuxhack = 0;
114 SYSCTL_INT(_vfs_nfsd, OID_AUTO, flexlinuxhack, CTLFLAG_RW,
115     &nfsrv_flexlinuxhack, 0,
116     "For Linux clients, hack around Flex File Layout bug");
117 
118 /*
119  * Hash lists for nfs V4.
120  */
121 NFSD_VNET_DEFINE(struct nfsclienthashhead *, nfsclienthash);
122 NFSD_VNET_DEFINE(struct nfslockhashhead *, nfslockhash);
123 NFSD_VNET_DEFINE(struct nfssessionhash *, nfssessionhash);
124 
125 struct nfslayouthash		*nfslayouthash;
126 volatile int nfsrv_dontlistlen = 0;
127 
128 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
129 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
130 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
131 static int nfsrv_nogsscallback = 0;
132 static volatile int nfsrv_writedelegcnt = 0;
133 static int nfsrv_faildscnt;
134 
135 NFSD_VNET_DEFINE_STATIC(time_t, nfsrvboottime);
136 
137 /* local functions */
138 static void nfsrv_dumpaclient(struct nfsclient *clp,
139     struct nfsd_dumpclients *dumpp);
140 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
141     NFSPROC_T *p);
142 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
143     NFSPROC_T *p);
144 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
145     NFSPROC_T *p);
146 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
147     int cansleep, NFSPROC_T *p);
148 static void nfsrv_freenfslock(struct nfslock *lop);
149 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
150 static void nfsrv_freedeleg(struct nfsstate *);
151 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
152     u_int32_t flags, struct nfsstate **stpp);
153 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
154     struct nfsstate **stpp);
155 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
156     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
157 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
158     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
159 static void nfsrv_insertlock(struct nfslock *new_lop,
160     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
161 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
162     struct nfslock **other_lopp, struct nfslockfile *lfp);
163 static int nfsrv_getipnumber(u_char *cp);
164 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
165     nfsv4stateid_t *stateidp, int specialid);
166 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
167     u_int32_t flags);
168 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
169     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
170     struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p);
171 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
172     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
173     int *slotposp);
174 static u_int32_t nfsrv_nextclientindex(void);
175 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
176 static void nfsrv_markstable(struct nfsclient *clp);
177 static void nfsrv_markreclaim(struct nfsclient *clp);
178 static int nfsrv_checkstable(struct nfsclient *clp);
179 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
180     vnode *vp, NFSPROC_T *p);
181 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
182     NFSPROC_T *p, vnode_t vp);
183 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
184     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
185 static int nfsrv_notsamecredname(int op, struct nfsrv_descript *nd,
186     struct nfsclient *clp);
187 static time_t nfsrv_leaseexpiry(void);
188 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
189 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
190     struct nfsstate *stp, struct nfsrvcache *op);
191 static int nfsrv_nootherstate(struct nfsstate *stp);
192 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
193     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
194 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
195     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
196 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
197     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
198     NFSPROC_T *p);
199 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
200     NFSPROC_T *p);
201 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
202     uint64_t first, uint64_t end);
203 static void nfsrv_locklf(struct nfslockfile *lfp);
204 static void nfsrv_unlocklf(struct nfslockfile *lfp);
205 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
206 static int nfsrv_freesession(struct nfsrv_descript *nd, struct nfsdsession *sep,
207     uint8_t *sessionid);
208 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
209     int dont_replycache, struct nfsdsession **sepp, int *slotposp);
210 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
211 static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
212     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p);
213 static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp);
214 static void nfsrv_freelayoutlist(nfsquad_t clientid);
215 static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype,
216     int iomode);
217 static void nfsrv_freealllayouts(void);
218 static void nfsrv_freedevid(struct nfsdevice *ds);
219 static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
220     struct nfsdevice **dsp);
221 static void nfsrv_deleteds(struct nfsdevice *fndds);
222 static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost);
223 static void nfsrv_freealldevids(void);
224 static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp,
225     int maxcnt, NFSPROC_T *p);
226 static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp,
227     fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype,
228     NFSPROC_T *p);
229 static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
230     NFSPROC_T *, struct nfslayout **lypp);
231 static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt);
232 static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode,
233     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
234 static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode,
235     int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
236 static int nfsrv_dontlayout(fhandle_t *fhp);
237 static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
238     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
239     vnode_t *tvpp);
240 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp);
241 static int nfsrv_checkmachcred(int op, struct nfsrv_descript *nd,
242     struct nfsclient *clp);
243 
244 /*
245  * Scan the client list for a match and either return the current one,
246  * create a new entry or return an error.
247  * If returning a non-error, the clp structure must either be linked into
248  * the client list or free'd.
249  */
250 int
251 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
252     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
253 {
254 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
255 	int i, error = 0, ret;
256 	struct nfsstate *stp, *tstp;
257 #ifdef INET
258 	struct sockaddr_in *sin, *rin;
259 #endif
260 #ifdef INET6
261 	struct sockaddr_in6 *sin6, *rin6;
262 #endif
263 	struct nfsdsession *sep, *nsep;
264 	int zapit = 0, gotit, hasstate = 0, igotlock;
265 	static u_int64_t confirm_index = 0;
266 
267 	/*
268 	 * Check for state resource limit exceeded.
269 	 */
270 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
271 		error = NFSERR_RESOURCE;
272 		goto out;
273 	}
274 
275 	if (nfsrv_issuedelegs == 0 ||
276 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
277 		/*
278 		 * Don't do callbacks when delegations are disabled or
279 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
280 		 * If establishing a callback connection is attempted
281 		 * when a firewall is blocking the callback path, the
282 		 * server may wait too long for the connect attempt to
283 		 * succeed during the Open. Some clients, such as Linux,
284 		 * may timeout and give up on the Open before the server
285 		 * replies. Also, since AUTH_GSS callbacks are not
286 		 * yet interoperability tested, they might cause the
287 		 * server to crap out, if they get past the Init call to
288 		 * the client.
289 		 */
290 		new_clp->lc_program = 0;
291 
292 	/* Lock out other nfsd threads */
293 	NFSLOCKV4ROOTMUTEX();
294 	nfsv4_relref(&nfsv4rootfs_lock);
295 	do {
296 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
297 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
298 	} while (!igotlock);
299 	NFSUNLOCKV4ROOTMUTEX();
300 
301 	/*
302 	 * Search for a match in the client list.
303 	 */
304 	gotit = i = 0;
305 	while (i < nfsrv_clienthashsize && !gotit) {
306 	    LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
307 		if (new_clp->lc_idlen == clp->lc_idlen &&
308 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
309 			gotit = 1;
310 			break;
311 		}
312 	    }
313 	    if (gotit == 0)
314 		i++;
315 	}
316 	if (!gotit ||
317 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
318 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
319 			/*
320 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
321 			 * client is trying to update a confirmed clientid.
322 			 */
323 			NFSLOCKV4ROOTMUTEX();
324 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
325 			NFSUNLOCKV4ROOTMUTEX();
326 			confirmp->lval[1] = 0;
327 			error = NFSERR_NOENT;
328 			goto out;
329 		}
330 		/*
331 		 * Get rid of the old one.
332 		 */
333 		if (i != nfsrv_clienthashsize) {
334 			LIST_REMOVE(clp, lc_hash);
335 			nfsrv_cleanclient(clp, p);
336 			nfsrv_freedeleglist(&clp->lc_deleg);
337 			nfsrv_freedeleglist(&clp->lc_olddeleg);
338 			zapit = 1;
339 		}
340 		/*
341 		 * Add it after assigning a client id to it.
342 		 */
343 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
344 		if ((nd->nd_flag & ND_NFSV41) != 0) {
345 			confirmp->lval[0] = ++confirm_index;
346 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
347 		} else
348 			confirmp->qval = new_clp->lc_confirm.qval =
349 			    ++confirm_index;
350 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
351 		    NFSD_VNET(nfsrvboottime);
352 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
353 		    nfsrv_nextclientindex();
354 		new_clp->lc_stateindex = 0;
355 		new_clp->lc_statemaxindex = 0;
356 		new_clp->lc_prevsess = 0;
357 		new_clp->lc_cbref = 0;
358 		new_clp->lc_expiry = nfsrv_leaseexpiry();
359 		LIST_INIT(&new_clp->lc_open);
360 		LIST_INIT(&new_clp->lc_deleg);
361 		LIST_INIT(&new_clp->lc_olddeleg);
362 		LIST_INIT(&new_clp->lc_session);
363 		for (i = 0; i < nfsrv_statehashsize; i++)
364 			LIST_INIT(&new_clp->lc_stateid[i]);
365 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
366 		    lc_hash);
367 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
368 		nfsrv_openpluslock++;
369 		nfsrv_clients++;
370 		NFSLOCKV4ROOTMUTEX();
371 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
372 		NFSUNLOCKV4ROOTMUTEX();
373 		if (zapit)
374 			nfsrv_zapclient(clp, p);
375 		*new_clpp = NULL;
376 		goto out;
377 	}
378 
379 	/*
380 	 * Now, handle the cases where the id is already issued.
381 	 */
382 	if (nfsrv_notsamecredname(NFSV4OP_EXCHANGEID, nd, clp)) {
383 	    /*
384 	     * Check to see if there is expired state that should go away.
385 	     */
386 	    if (clp->lc_expiry < NFSD_MONOSEC &&
387 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
388 		nfsrv_cleanclient(clp, p);
389 		nfsrv_freedeleglist(&clp->lc_deleg);
390 	    }
391 
392 	    /*
393 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
394 	     * RFC3530 Sec. 8.1.2 last para.
395 	     */
396 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
397 		hasstate = 1;
398 	    } else if (LIST_EMPTY(&clp->lc_open)) {
399 		hasstate = 0;
400 	    } else {
401 		hasstate = 0;
402 		/* Look for an Open on the OpenOwner */
403 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
404 		    if (!LIST_EMPTY(&stp->ls_open)) {
405 			hasstate = 1;
406 			break;
407 		    }
408 		}
409 	    }
410 	    if (hasstate) {
411 		/*
412 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
413 		 * filling out the correct ipaddr and portnum.
414 		 */
415 		switch (clp->lc_req.nr_nam->sa_family) {
416 #ifdef INET
417 		case AF_INET:
418 			sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam;
419 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
420 			sin->sin_addr.s_addr = rin->sin_addr.s_addr;
421 			sin->sin_port = rin->sin_port;
422 			break;
423 #endif
424 #ifdef INET6
425 		case AF_INET6:
426 			sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam;
427 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
428 			sin6->sin6_addr = rin6->sin6_addr;
429 			sin6->sin6_port = rin6->sin6_port;
430 			break;
431 #endif
432 		}
433 		NFSLOCKV4ROOTMUTEX();
434 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
435 		NFSUNLOCKV4ROOTMUTEX();
436 		error = NFSERR_CLIDINUSE;
437 		goto out;
438 	    }
439 	}
440 
441 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
442 		/*
443 		 * If the verifier has changed, the client has rebooted
444 		 * and a new client id is issued. The old state info
445 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
446 		 */
447 		LIST_REMOVE(clp, lc_hash);
448 
449 		/* Get rid of all sessions on this clientid. */
450 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) {
451 			ret = nfsrv_freesession(NULL, sep, NULL);
452 			if (ret != 0)
453 				printf("nfsrv_setclient: verifier changed free"
454 				    " session failed=%d\n", ret);
455 		}
456 
457 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
458 		if ((nd->nd_flag & ND_NFSV41) != 0) {
459 			confirmp->lval[0] = ++confirm_index;
460 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
461 		} else
462 			confirmp->qval = new_clp->lc_confirm.qval =
463 			    ++confirm_index;
464 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
465 		    NFSD_VNET(nfsrvboottime);
466 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
467 		    nfsrv_nextclientindex();
468 		new_clp->lc_stateindex = 0;
469 		new_clp->lc_statemaxindex = 0;
470 		new_clp->lc_prevsess = 0;
471 		new_clp->lc_cbref = 0;
472 		new_clp->lc_expiry = nfsrv_leaseexpiry();
473 
474 		/*
475 		 * Save the state until confirmed.
476 		 */
477 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
478 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
479 			tstp->ls_clp = new_clp;
480 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
481 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
482 			tstp->ls_clp = new_clp;
483 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
484 		    ls_list);
485 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
486 			tstp->ls_clp = new_clp;
487 		for (i = 0; i < nfsrv_statehashsize; i++) {
488 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
489 			    &clp->lc_stateid[i], ls_hash);
490 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
491 				tstp->ls_clp = new_clp;
492 		}
493 		LIST_INIT(&new_clp->lc_session);
494 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
495 		    lc_hash);
496 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
497 		nfsrv_openpluslock++;
498 		nfsrv_clients++;
499 		NFSLOCKV4ROOTMUTEX();
500 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
501 		NFSUNLOCKV4ROOTMUTEX();
502 
503 		/*
504 		 * Must wait until any outstanding callback on the old clp
505 		 * completes.
506 		 */
507 		NFSLOCKSTATE();
508 		while (clp->lc_cbref) {
509 			clp->lc_flags |= LCL_WAKEUPWANTED;
510 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
511 			    "nfsd clp", 10 * hz);
512 		}
513 		NFSUNLOCKSTATE();
514 		nfsrv_zapclient(clp, p);
515 		*new_clpp = NULL;
516 		goto out;
517 	}
518 
519 	/* For NFSv4.1, mark that we found a confirmed clientid. */
520 	if ((nd->nd_flag & ND_NFSV41) != 0) {
521 		clientidp->lval[0] = clp->lc_clientid.lval[0];
522 		clientidp->lval[1] = clp->lc_clientid.lval[1];
523 		confirmp->lval[0] = 0;	/* Ignored by client */
524 		confirmp->lval[1] = 1;
525 	} else {
526 		/*
527 		 * id and verifier match, so update the net address info
528 		 * and get rid of any existing callback authentication
529 		 * handle, so a new one will be acquired.
530 		 */
531 		LIST_REMOVE(clp, lc_hash);
532 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
533 		new_clp->lc_expiry = nfsrv_leaseexpiry();
534 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
535 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
536 		    clp->lc_clientid.lval[0];
537 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
538 		    clp->lc_clientid.lval[1];
539 		new_clp->lc_delegtime = clp->lc_delegtime;
540 		new_clp->lc_stateindex = clp->lc_stateindex;
541 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
542 		new_clp->lc_cbref = 0;
543 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
544 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
545 			tstp->ls_clp = new_clp;
546 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
547 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
548 			tstp->ls_clp = new_clp;
549 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
550 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
551 			tstp->ls_clp = new_clp;
552 		for (i = 0; i < nfsrv_statehashsize; i++) {
553 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
554 			    &clp->lc_stateid[i], ls_hash);
555 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
556 				tstp->ls_clp = new_clp;
557 		}
558 		LIST_INIT(&new_clp->lc_session);
559 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
560 		    lc_hash);
561 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
562 		nfsrv_openpluslock++;
563 		nfsrv_clients++;
564 	}
565 	NFSLOCKV4ROOTMUTEX();
566 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
567 	NFSUNLOCKV4ROOTMUTEX();
568 
569 	if ((nd->nd_flag & ND_NFSV41) == 0) {
570 		/*
571 		 * Must wait until any outstanding callback on the old clp
572 		 * completes.
573 		 */
574 		NFSLOCKSTATE();
575 		while (clp->lc_cbref) {
576 			clp->lc_flags |= LCL_WAKEUPWANTED;
577 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
578 			    "nfsdclp", 10 * hz);
579 		}
580 		NFSUNLOCKSTATE();
581 		nfsrv_zapclient(clp, p);
582 		*new_clpp = NULL;
583 	}
584 
585 out:
586 	NFSEXITCODE2(error, nd);
587 	return (error);
588 }
589 
590 /*
591  * Check to see if the client id exists and optionally confirm it.
592  */
593 int
594 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
595     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
596     struct nfsrv_descript *nd, NFSPROC_T *p)
597 {
598 	struct nfsclient *clp;
599 	struct nfsstate *stp;
600 	int i;
601 	struct nfsclienthashhead *hp;
602 	int error = 0, igotlock, doneok;
603 	struct nfssessionhash *shp;
604 	struct nfsdsession *sep;
605 	uint64_t sessid[2];
606 	bool sess_replay;
607 	static uint64_t next_sess = 0;
608 
609 	if (clpp)
610 		*clpp = NULL;
611 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
612 	    opflags != CLOPS_RENEW) && NFSD_VNET(nfsrvboottime) !=
613 	    clientid.lval[0]) {
614 		error = NFSERR_STALECLIENTID;
615 		goto out;
616 	}
617 
618 	/*
619 	 * If called with opflags == CLOPS_RENEW, the State Lock is
620 	 * already held. Otherwise, we need to get either that or,
621 	 * for the case of Confirm, lock out the nfsd threads.
622 	 */
623 	if (opflags & CLOPS_CONFIRM) {
624 		NFSLOCKV4ROOTMUTEX();
625 		nfsv4_relref(&nfsv4rootfs_lock);
626 		do {
627 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
628 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
629 		} while (!igotlock);
630 		/*
631 		 * Create a new sessionid here, since we need to do it where
632 		 * there is a mutex held to serialize update of next_sess.
633 		 */
634 		if ((nd->nd_flag & ND_NFSV41) != 0) {
635 			sessid[0] = ++next_sess;
636 			sessid[1] = clientid.qval;
637 		}
638 		NFSUNLOCKV4ROOTMUTEX();
639 	} else if (opflags != CLOPS_RENEW) {
640 		NFSLOCKSTATE();
641 	}
642 
643 	/* For NFSv4.1, the clp is acquired from the associated session. */
644 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
645 	    opflags == CLOPS_RENEW) {
646 		clp = NULL;
647 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
648 			shp = NFSSESSIONHASH(nd->nd_sessionid);
649 			NFSLOCKSESSION(shp);
650 			sep = nfsrv_findsession(nd->nd_sessionid);
651 			if (sep != NULL)
652 				clp = sep->sess_clp;
653 			NFSUNLOCKSESSION(shp);
654 		}
655 	} else {
656 		hp = NFSCLIENTHASH(clientid);
657 		LIST_FOREACH(clp, hp, lc_hash) {
658 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
659 				break;
660 		}
661 	}
662 	if (clp == NULL) {
663 		if (opflags & CLOPS_CONFIRM)
664 			error = NFSERR_STALECLIENTID;
665 		else
666 			error = NFSERR_EXPIRED;
667 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
668 		/*
669 		 * If marked admin revoked, just return the error.
670 		 */
671 		error = NFSERR_ADMINREVOKED;
672 	}
673 	if (error) {
674 		if (opflags & CLOPS_CONFIRM) {
675 			NFSLOCKV4ROOTMUTEX();
676 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
677 			NFSUNLOCKV4ROOTMUTEX();
678 		} else if (opflags != CLOPS_RENEW) {
679 			NFSUNLOCKSTATE();
680 		}
681 		goto out;
682 	}
683 
684 	/*
685 	 * Perform any operations specified by the opflags.
686 	 */
687 	if (opflags & CLOPS_CONFIRM) {
688 		sess_replay = false;
689 		if ((nd->nd_flag & ND_NFSV41) != 0) {
690 		    /*
691 		     * For the case where lc_confirm.lval[0] == confirm.lval[0],
692 		     * use the new session, but with the previous sessionid.
693 		     * This is not exactly what the RFC describes, but should
694 		     * result in the same reply as the previous CreateSession.
695 		     */
696 		    if (clp->lc_confirm.lval[0] + 1 == confirm.lval[0]) {
697 			clp->lc_confirm.lval[0] = confirm.lval[0];
698 			clp->lc_prevsess = sessid[0];
699 		    } else if (clp->lc_confirm.lval[0] == confirm.lval[0]) {
700 			if (clp->lc_prevsess == 0)
701 			    error = NFSERR_SEQMISORDERED;
702 			else
703 			    sessid[0] = clp->lc_prevsess;
704 			sess_replay = true;
705 		    } else
706 			error = NFSERR_SEQMISORDERED;
707 		} else if ((nd->nd_flag & ND_NFSV41) == 0 &&
708 		     clp->lc_confirm.qval != confirm.qval)
709 			error = NFSERR_STALECLIENTID;
710 		if (error == 0 && nfsrv_notsamecredname(NFSV4OP_CREATESESSION,
711 		    nd, clp))
712 			error = NFSERR_CLIDINUSE;
713 
714 		if (!error) {
715 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
716 			LCL_NEEDSCONFIRM) {
717 			/*
718 			 * Hang onto the delegations (as old delegations)
719 			 * for an Open with CLAIM_DELEGATE_PREV unless in
720 			 * grace, but get rid of the rest of the state.
721 			 */
722 			nfsrv_cleanclient(clp, p);
723 			nfsrv_freedeleglist(&clp->lc_olddeleg);
724 			if (nfsrv_checkgrace(nd, clp, 0)) {
725 			    /* In grace, so just delete delegations */
726 			    nfsrv_freedeleglist(&clp->lc_deleg);
727 			} else {
728 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
729 				stp->ls_flags |= NFSLCK_OLDDELEG;
730 			    clp->lc_delegtime = NFSD_MONOSEC +
731 				nfsrv_lease + NFSRV_LEASEDELTA;
732 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
733 				ls_list);
734 			}
735 			if ((nd->nd_flag & ND_NFSV41) != 0)
736 			    clp->lc_program = cbprogram;
737 		    }
738 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
739 		    if (clp->lc_program)
740 			clp->lc_flags |= LCL_NEEDSCBNULL;
741 		    /* For NFSv4.1, link the session onto the client. */
742 		    if (nsep != NULL) {
743 			/* Hold a reference on the xprt for a backchannel. */
744 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
745 			    != 0 && !sess_replay) {
746 			    if (clp->lc_req.nr_client == NULL)
747 				clp->lc_req.nr_client = (struct __rpc_client *)
748 				    clnt_bck_create(nd->nd_xprt->xp_socket,
749 				    cbprogram, NFSV4_CBVERS);
750 			    if (clp->lc_req.nr_client != NULL) {
751 				SVC_ACQUIRE(nd->nd_xprt);
752 				CLNT_ACQUIRE(clp->lc_req.nr_client);
753 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
754 				/* Disable idle timeout. */
755 				nd->nd_xprt->xp_idletimeout = 0;
756 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
757 			    } else
758 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
759 			}
760 			NFSBCOPY(sessid, nsep->sess_sessionid,
761 			    NFSX_V4SESSIONID);
762 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
763 			    NFSX_V4SESSIONID);
764 			if (!sess_replay) {
765 			    shp = NFSSESSIONHASH(nsep->sess_sessionid);
766 			    NFSLOCKSTATE();
767 			    NFSLOCKSESSION(shp);
768 			    LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
769 			    LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
770 			    nsep->sess_clp = clp;
771 			    NFSUNLOCKSESSION(shp);
772 			    NFSUNLOCKSTATE();
773 			}
774 		    }
775 		}
776 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
777 		error = NFSERR_EXPIRED;
778 	}
779 
780 	/*
781 	 * If called by the Renew Op, we must check the principal.
782 	 */
783 	if (!error && (opflags & CLOPS_RENEWOP)) {
784 	    if (nfsrv_notsamecredname(0, nd, clp)) {
785 		doneok = 0;
786 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
787 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
788 			if ((stp->ls_flags & NFSLCK_OPEN) &&
789 			    stp->ls_uid == nd->nd_cred->cr_uid) {
790 				doneok = 1;
791 				break;
792 			}
793 		    }
794 		}
795 		if (!doneok)
796 			error = NFSERR_ACCES;
797 	    }
798 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
799 		error = NFSERR_CBPATHDOWN;
800 	}
801 	if ((!error || error == NFSERR_CBPATHDOWN) &&
802 	     (opflags & CLOPS_RENEW)) {
803 		clp->lc_expiry = nfsrv_leaseexpiry();
804 	}
805 	if (opflags & CLOPS_CONFIRM) {
806 		NFSLOCKV4ROOTMUTEX();
807 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
808 		NFSUNLOCKV4ROOTMUTEX();
809 	} else if (opflags != CLOPS_RENEW) {
810 		NFSUNLOCKSTATE();
811 	}
812 	if (clpp)
813 		*clpp = clp;
814 
815 out:
816 	NFSEXITCODE2(error, nd);
817 	return (error);
818 }
819 
820 /*
821  * Perform the NFSv4.1 destroy clientid.
822  */
823 int
824 nfsrv_destroyclient(struct nfsrv_descript *nd, nfsquad_t clientid, NFSPROC_T *p)
825 {
826 	struct nfsclient *clp;
827 	struct nfsclienthashhead *hp;
828 	int error = 0, i, igotlock;
829 
830 	if (NFSD_VNET(nfsrvboottime) != clientid.lval[0]) {
831 		error = NFSERR_STALECLIENTID;
832 		goto out;
833 	}
834 
835 	/* Lock out other nfsd threads */
836 	NFSLOCKV4ROOTMUTEX();
837 	nfsv4_relref(&nfsv4rootfs_lock);
838 	do {
839 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
840 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
841 	} while (igotlock == 0);
842 	NFSUNLOCKV4ROOTMUTEX();
843 
844 	hp = NFSCLIENTHASH(clientid);
845 	LIST_FOREACH(clp, hp, lc_hash) {
846 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
847 			break;
848 	}
849 	if (clp == NULL) {
850 		NFSLOCKV4ROOTMUTEX();
851 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
852 		NFSUNLOCKV4ROOTMUTEX();
853 		/* Just return ok, since it is gone. */
854 		goto out;
855 	}
856 
857 	/* Check for the SP4_MACH_CRED case. */
858 	error = nfsrv_checkmachcred(NFSV4OP_DESTROYCLIENTID, nd, clp);
859 	if (error != 0) {
860 		NFSLOCKV4ROOTMUTEX();
861 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
862 		NFSUNLOCKV4ROOTMUTEX();
863 		goto out;
864 	}
865 
866 	/*
867 	 * Free up all layouts on the clientid.  Should the client return the
868 	 * layouts?
869 	 */
870 	nfsrv_freelayoutlist(clientid);
871 
872 	/* Scan for state on the clientid. */
873 	for (i = 0; i < nfsrv_statehashsize; i++)
874 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
875 			NFSLOCKV4ROOTMUTEX();
876 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
877 			NFSUNLOCKV4ROOTMUTEX();
878 			error = NFSERR_CLIENTIDBUSY;
879 			goto out;
880 		}
881 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
882 		NFSLOCKV4ROOTMUTEX();
883 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
884 		NFSUNLOCKV4ROOTMUTEX();
885 		error = NFSERR_CLIENTIDBUSY;
886 		goto out;
887 	}
888 
889 	/* Destroy the clientid and return ok. */
890 	nfsrv_cleanclient(clp, p);
891 	nfsrv_freedeleglist(&clp->lc_deleg);
892 	nfsrv_freedeleglist(&clp->lc_olddeleg);
893 	LIST_REMOVE(clp, lc_hash);
894 	NFSLOCKV4ROOTMUTEX();
895 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
896 	NFSUNLOCKV4ROOTMUTEX();
897 	nfsrv_zapclient(clp, p);
898 out:
899 	NFSEXITCODE2(error, nd);
900 	return (error);
901 }
902 
903 /*
904  * Called from the new nfssvc syscall to admin revoke a clientid.
905  * Returns 0 for success, error otherwise.
906  */
907 int
908 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
909 {
910 	struct nfsclient *clp = NULL;
911 	int i, error = 0;
912 	int gotit, igotlock;
913 
914 	/*
915 	 * First, lock out the nfsd so that state won't change while the
916 	 * revocation record is being written to the stable storage restart
917 	 * file.
918 	 */
919 	NFSLOCKV4ROOTMUTEX();
920 	do {
921 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
922 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
923 	} while (!igotlock);
924 	NFSUNLOCKV4ROOTMUTEX();
925 
926 	/*
927 	 * Search for a match in the client list.
928 	 */
929 	gotit = i = 0;
930 	while (i < nfsrv_clienthashsize && !gotit) {
931 	    LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
932 		if (revokep->nclid_idlen == clp->lc_idlen &&
933 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
934 			gotit = 1;
935 			break;
936 		}
937 	    }
938 	    i++;
939 	}
940 	if (!gotit) {
941 		NFSLOCKV4ROOTMUTEX();
942 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
943 		NFSUNLOCKV4ROOTMUTEX();
944 		error = EPERM;
945 		goto out;
946 	}
947 
948 	/*
949 	 * Now, write out the revocation record
950 	 */
951 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
952 	nfsrv_backupstable();
953 
954 	/*
955 	 * and clear out the state, marking the clientid revoked.
956 	 */
957 	clp->lc_flags &= ~LCL_CALLBACKSON;
958 	clp->lc_flags |= LCL_ADMINREVOKED;
959 	nfsrv_cleanclient(clp, p);
960 	nfsrv_freedeleglist(&clp->lc_deleg);
961 	nfsrv_freedeleglist(&clp->lc_olddeleg);
962 	NFSLOCKV4ROOTMUTEX();
963 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
964 	NFSUNLOCKV4ROOTMUTEX();
965 
966 out:
967 	NFSEXITCODE(error);
968 	return (error);
969 }
970 
971 /*
972  * Dump out stats for all clients. Called from nfssvc(2), that is used
973  * nfsstatsv1.
974  */
975 void
976 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
977 {
978 	struct nfsclient *clp;
979 	int i = 0, cnt = 0;
980 
981 	/*
982 	 * First, get a reference on the nfsv4rootfs_lock so that an
983 	 * exclusive lock cannot be acquired while dumping the clients.
984 	 */
985 	NFSLOCKV4ROOTMUTEX();
986 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
987 	NFSUNLOCKV4ROOTMUTEX();
988 	NFSLOCKSTATE();
989 	/*
990 	 * Rattle through the client lists until done.
991 	 */
992 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
993 	    clp = LIST_FIRST(&NFSD_VNET(nfsclienthash)[i]);
994 	    while (clp != LIST_END(&NFSD_VNET(nfsclienthash)[i]) && cnt <
995 		maxcnt) {
996 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
997 		cnt++;
998 		clp = LIST_NEXT(clp, lc_hash);
999 	    }
1000 	    i++;
1001 	}
1002 	if (cnt < maxcnt)
1003 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
1004 	NFSUNLOCKSTATE();
1005 	NFSLOCKV4ROOTMUTEX();
1006 	nfsv4_relref(&nfsv4rootfs_lock);
1007 	NFSUNLOCKV4ROOTMUTEX();
1008 }
1009 
1010 /*
1011  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
1012  */
1013 static void
1014 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
1015 {
1016 	struct nfsstate *stp, *openstp, *lckownstp;
1017 	struct nfslock *lop;
1018 	sa_family_t af;
1019 #ifdef INET
1020 	struct sockaddr_in *rin;
1021 #endif
1022 #ifdef INET6
1023 	struct sockaddr_in6 *rin6;
1024 #endif
1025 
1026 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
1027 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
1028 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
1029 	dumpp->ndcl_flags = clp->lc_flags;
1030 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
1031 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
1032 	af = clp->lc_req.nr_nam->sa_family;
1033 	dumpp->ndcl_addrfam = af;
1034 	switch (af) {
1035 #ifdef INET
1036 	case AF_INET:
1037 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
1038 		dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr;
1039 		break;
1040 #endif
1041 #ifdef INET6
1042 	case AF_INET6:
1043 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
1044 		dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr;
1045 		break;
1046 #endif
1047 	}
1048 
1049 	/*
1050 	 * Now, scan the state lists and total up the opens and locks.
1051 	 */
1052 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
1053 	    dumpp->ndcl_nopenowners++;
1054 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
1055 		dumpp->ndcl_nopens++;
1056 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
1057 		    dumpp->ndcl_nlockowners++;
1058 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
1059 			dumpp->ndcl_nlocks++;
1060 		    }
1061 		}
1062 	    }
1063 	}
1064 
1065 	/*
1066 	 * and the delegation lists.
1067 	 */
1068 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
1069 	    dumpp->ndcl_ndelegs++;
1070 	}
1071 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
1072 	    dumpp->ndcl_nolddelegs++;
1073 	}
1074 }
1075 
1076 /*
1077  * Dump out lock stats for a file.
1078  */
1079 void
1080 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
1081     NFSPROC_T *p)
1082 {
1083 	struct nfsstate *stp;
1084 	struct nfslock *lop;
1085 	int cnt = 0;
1086 	struct nfslockfile *lfp;
1087 	sa_family_t af;
1088 #ifdef INET
1089 	struct sockaddr_in *rin;
1090 #endif
1091 #ifdef INET6
1092 	struct sockaddr_in6 *rin6;
1093 #endif
1094 	int ret;
1095 	fhandle_t nfh;
1096 
1097 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
1098 	/*
1099 	 * First, get a reference on the nfsv4rootfs_lock so that an
1100 	 * exclusive lock on it cannot be acquired while dumping the locks.
1101 	 */
1102 	NFSLOCKV4ROOTMUTEX();
1103 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1104 	NFSUNLOCKV4ROOTMUTEX();
1105 	NFSLOCKSTATE();
1106 	if (!ret)
1107 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1108 	if (ret) {
1109 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
1110 		NFSUNLOCKSTATE();
1111 		NFSLOCKV4ROOTMUTEX();
1112 		nfsv4_relref(&nfsv4rootfs_lock);
1113 		NFSUNLOCKV4ROOTMUTEX();
1114 		return;
1115 	}
1116 
1117 	/*
1118 	 * For each open share on file, dump it out.
1119 	 */
1120 	stp = LIST_FIRST(&lfp->lf_open);
1121 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1122 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1123 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1124 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1125 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1126 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1127 		ldumpp[cnt].ndlck_owner.nclid_idlen =
1128 		    stp->ls_openowner->ls_ownerlen;
1129 		NFSBCOPY(stp->ls_openowner->ls_owner,
1130 		    ldumpp[cnt].ndlck_owner.nclid_id,
1131 		    stp->ls_openowner->ls_ownerlen);
1132 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1133 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1134 		    stp->ls_clp->lc_idlen);
1135 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1136 		ldumpp[cnt].ndlck_addrfam = af;
1137 		switch (af) {
1138 #ifdef INET
1139 		case AF_INET:
1140 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1141 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1142 			break;
1143 #endif
1144 #ifdef INET6
1145 		case AF_INET6:
1146 			rin6 = (struct sockaddr_in6 *)
1147 			    stp->ls_clp->lc_req.nr_nam;
1148 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1149 			break;
1150 #endif
1151 		}
1152 		stp = LIST_NEXT(stp, ls_file);
1153 		cnt++;
1154 	}
1155 
1156 	/*
1157 	 * and all locks.
1158 	 */
1159 	lop = LIST_FIRST(&lfp->lf_lock);
1160 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1161 		stp = lop->lo_stp;
1162 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1163 		ldumpp[cnt].ndlck_first = lop->lo_first;
1164 		ldumpp[cnt].ndlck_end = lop->lo_end;
1165 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1166 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1167 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1168 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1169 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1170 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1171 		    stp->ls_ownerlen);
1172 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1173 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1174 		    stp->ls_clp->lc_idlen);
1175 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1176 		ldumpp[cnt].ndlck_addrfam = af;
1177 		switch (af) {
1178 #ifdef INET
1179 		case AF_INET:
1180 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1181 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1182 			break;
1183 #endif
1184 #ifdef INET6
1185 		case AF_INET6:
1186 			rin6 = (struct sockaddr_in6 *)
1187 			    stp->ls_clp->lc_req.nr_nam;
1188 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1189 			break;
1190 #endif
1191 		}
1192 		lop = LIST_NEXT(lop, lo_lckfile);
1193 		cnt++;
1194 	}
1195 
1196 	/*
1197 	 * and the delegations.
1198 	 */
1199 	stp = LIST_FIRST(&lfp->lf_deleg);
1200 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1201 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1202 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1203 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1204 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1205 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1206 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1207 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1208 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1209 		    stp->ls_clp->lc_idlen);
1210 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1211 		ldumpp[cnt].ndlck_addrfam = af;
1212 		switch (af) {
1213 #ifdef INET
1214 		case AF_INET:
1215 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1216 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1217 			break;
1218 #endif
1219 #ifdef INET6
1220 		case AF_INET6:
1221 			rin6 = (struct sockaddr_in6 *)
1222 			    stp->ls_clp->lc_req.nr_nam;
1223 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1224 			break;
1225 #endif
1226 		}
1227 		stp = LIST_NEXT(stp, ls_file);
1228 		cnt++;
1229 	}
1230 
1231 	/*
1232 	 * If list isn't full, mark end of list by setting the client name
1233 	 * to zero length.
1234 	 */
1235 	if (cnt < maxcnt)
1236 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1237 	NFSUNLOCKSTATE();
1238 	NFSLOCKV4ROOTMUTEX();
1239 	nfsv4_relref(&nfsv4rootfs_lock);
1240 	NFSUNLOCKV4ROOTMUTEX();
1241 }
1242 
1243 /*
1244  * Server timer routine. It can scan any linked list, so long
1245  * as it holds the spin/mutex lock and there is no exclusive lock on
1246  * nfsv4rootfs_lock.
1247  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1248  *  to do this from a callout, since the spin locks work. For
1249  *  Darwin, I'm not sure what will work correctly yet.)
1250  * Should be called once per second.
1251  */
1252 void
1253 nfsrv_servertimer(void *arg __unused)
1254 {
1255 	struct nfsclient *clp, *nclp;
1256 	struct nfsstate *stp, *nstp;
1257 	int got_ref, i;
1258 
1259 	/*
1260 	 * Make sure nfsboottime is set. This is used by V3 as well
1261 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1262 	 * only used by the V4 server for leases.
1263 	 */
1264 	if (nfsboottime.tv_sec == 0)
1265 		NFSSETBOOTTIME(nfsboottime);
1266 
1267 	/*
1268 	 * If server hasn't started yet, just return.
1269 	 */
1270 	NFSLOCKSTATE();
1271 	if (NFSD_VNET(nfsrv_stablefirst).nsf_eograce == 0) {
1272 		NFSUNLOCKSTATE();
1273 		return;
1274 	}
1275 	if (!(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE)) {
1276 		if (!(NFSD_VNET(nfsrv_stablefirst).nsf_flags &
1277 		      NFSNSF_GRACEOVER) &&
1278 		    NFSD_MONOSEC > NFSD_VNET(nfsrv_stablefirst).nsf_eograce)
1279 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1280 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1281 		NFSUNLOCKSTATE();
1282 		return;
1283 	}
1284 
1285 	/*
1286 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1287 	 * no nfsd thread can acquire an exclusive lock on it before this
1288 	 * call is done. If it is already exclusively locked, just return.
1289 	 */
1290 	NFSLOCKV4ROOTMUTEX();
1291 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1292 	NFSUNLOCKV4ROOTMUTEX();
1293 	if (got_ref == 0) {
1294 		NFSUNLOCKSTATE();
1295 		return;
1296 	}
1297 
1298 	/*
1299 	 * For each client...
1300 	 */
1301 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1302 	    clp = LIST_FIRST(&NFSD_VNET(nfsclienthash)[i]);
1303 	    while (clp != LIST_END(&NFSD_VNET(nfsclienthash)[i])) {
1304 		nclp = LIST_NEXT(clp, lc_hash);
1305 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1306 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1307 			 && ((LIST_EMPTY(&clp->lc_deleg)
1308 			      && LIST_EMPTY(&clp->lc_open)) ||
1309 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1310 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1311 			(clp->lc_expiry < NFSD_MONOSEC &&
1312 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1313 			/*
1314 			 * Lease has expired several nfsrv_lease times ago:
1315 			 * PLUS
1316 			 *    - no state is associated with it
1317 			 *    OR
1318 			 *    - above high water mark for number of clients
1319 			 *      (nfsrv_clienthighwater should be large enough
1320 			 *       that this only occurs when clients fail to
1321 			 *       use the same nfs_client_id4.id. Maybe somewhat
1322 			 *       higher that the maximum number of clients that
1323 			 *       will mount this server?)
1324 			 * OR
1325 			 * Lease has expired a very long time ago
1326 			 * OR
1327 			 * Lease has expired PLUS the number of opens + locks
1328 			 * has exceeded 90% of capacity
1329 			 *
1330 			 * --> Mark for expiry. The actual expiry will be done
1331 			 *     by an nfsd sometime soon.
1332 			 */
1333 			clp->lc_flags |= LCL_EXPIREIT;
1334 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1335 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1336 		    } else {
1337 			/*
1338 			 * If there are no opens, increment no open tick cnt
1339 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1340 			 * otherwise, if there is an open, reset no open time
1341 			 * Hopefully, this will avoid excessive re-creation
1342 			 * of open owners and subsequent open confirms.
1343 			 */
1344 			stp = LIST_FIRST(&clp->lc_open);
1345 			while (stp != LIST_END(&clp->lc_open)) {
1346 				nstp = LIST_NEXT(stp, ls_list);
1347 				if (LIST_EMPTY(&stp->ls_open)) {
1348 					stp->ls_noopens++;
1349 					if (stp->ls_noopens > NFSNOOPEN ||
1350 					    (nfsrv_openpluslock * 2) >
1351 					    nfsrv_v4statelimit)
1352 						NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1353 							NFSNSF_NOOPENS;
1354 				} else {
1355 					stp->ls_noopens = 0;
1356 				}
1357 				stp = nstp;
1358 			}
1359 		    }
1360 		}
1361 		clp = nclp;
1362 	    }
1363 	}
1364 	NFSUNLOCKSTATE();
1365 	NFSLOCKV4ROOTMUTEX();
1366 	nfsv4_relref(&nfsv4rootfs_lock);
1367 	NFSUNLOCKV4ROOTMUTEX();
1368 }
1369 
1370 /*
1371  * The following set of functions free up the various data structures.
1372  */
1373 /*
1374  * Clear out all open/lock state related to this nfsclient.
1375  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1376  * there are no other active nfsd threads.
1377  */
1378 void
1379 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1380 {
1381 	struct nfsstate *stp, *nstp;
1382 	struct nfsdsession *sep, *nsep;
1383 
1384 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1385 		nfsrv_freeopenowner(stp, 1, p);
1386 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1387 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1388 			(void)nfsrv_freesession(NULL, sep, NULL);
1389 }
1390 
1391 /*
1392  * Free a client that has been cleaned. It should also already have been
1393  * removed from the lists.
1394  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1395  *  softclock interrupts are enabled.)
1396  */
1397 void
1398 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1399 {
1400 
1401 #ifdef notyet
1402 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1403 	     (LCL_GSS | LCL_CALLBACKSON) &&
1404 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1405 	    clp->lc_handlelen > 0) {
1406 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1407 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1408 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1409 			NULL, 0, NULL, NULL, NULL, 0, p);
1410 	}
1411 #endif
1412 	newnfs_disconnect(NULL, &clp->lc_req);
1413 	free(clp->lc_req.nr_nam, M_SONAME);
1414 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1415 	free(clp->lc_stateid, M_NFSDCLIENT);
1416 	free(clp, M_NFSDCLIENT);
1417 	NFSLOCKSTATE();
1418 	NFSD_VNET(nfsstatsv1_p)->srvclients--;
1419 	nfsrv_openpluslock--;
1420 	nfsrv_clients--;
1421 	NFSUNLOCKSTATE();
1422 }
1423 
1424 /*
1425  * Free a list of delegation state structures.
1426  * (This function will also free all nfslockfile structures that no
1427  *  longer have associated state.)
1428  */
1429 void
1430 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1431 {
1432 	struct nfsstate *stp, *nstp;
1433 
1434 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1435 		nfsrv_freedeleg(stp);
1436 	}
1437 	LIST_INIT(sthp);
1438 }
1439 
1440 /*
1441  * Free up a delegation.
1442  */
1443 static void
1444 nfsrv_freedeleg(struct nfsstate *stp)
1445 {
1446 	struct nfslockfile *lfp;
1447 
1448 	LIST_REMOVE(stp, ls_hash);
1449 	LIST_REMOVE(stp, ls_list);
1450 	LIST_REMOVE(stp, ls_file);
1451 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1452 		nfsrv_writedelegcnt--;
1453 	lfp = stp->ls_lfp;
1454 	if (LIST_EMPTY(&lfp->lf_open) &&
1455 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1456 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1457 	    lfp->lf_usecount == 0 &&
1458 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1459 		nfsrv_freenfslockfile(lfp);
1460 	free(stp, M_NFSDSTATE);
1461 	NFSD_VNET(nfsstatsv1_p)->srvdelegates--;
1462 	nfsrv_openpluslock--;
1463 	nfsrv_delegatecnt--;
1464 }
1465 
1466 /*
1467  * This function frees an open owner and all associated opens.
1468  */
1469 static void
1470 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1471 {
1472 	struct nfsstate *nstp, *tstp;
1473 
1474 	LIST_REMOVE(stp, ls_list);
1475 	/*
1476 	 * Now, free all associated opens.
1477 	 */
1478 	nstp = LIST_FIRST(&stp->ls_open);
1479 	while (nstp != LIST_END(&stp->ls_open)) {
1480 		tstp = nstp;
1481 		nstp = LIST_NEXT(nstp, ls_list);
1482 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1483 	}
1484 	if (stp->ls_op)
1485 		nfsrvd_derefcache(stp->ls_op);
1486 	free(stp, M_NFSDSTATE);
1487 	NFSD_VNET(nfsstatsv1_p)->srvopenowners--;
1488 	nfsrv_openpluslock--;
1489 }
1490 
1491 /*
1492  * This function frees an open (nfsstate open structure) with all associated
1493  * lock_owners and locks. It also frees the nfslockfile structure iff there
1494  * are no other opens on the file.
1495  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1496  */
1497 static int
1498 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1499 {
1500 	struct nfsstate *nstp, *tstp;
1501 	struct nfslockfile *lfp;
1502 	int ret;
1503 
1504 	LIST_REMOVE(stp, ls_hash);
1505 	LIST_REMOVE(stp, ls_list);
1506 	LIST_REMOVE(stp, ls_file);
1507 
1508 	lfp = stp->ls_lfp;
1509 	/*
1510 	 * Now, free all lockowners associated with this open.
1511 	 */
1512 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1513 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1514 
1515 	/*
1516 	 * The nfslockfile is freed here if there are no locks
1517 	 * associated with the open.
1518 	 * If there are locks associated with the open, the
1519 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1520 	 * Acquire the state mutex to avoid races with calls to
1521 	 * nfsrv_getlockfile().
1522 	 */
1523 	if (cansleep != 0)
1524 		NFSLOCKSTATE();
1525 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1526 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1527 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1528 	    lfp->lf_usecount == 0 &&
1529 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1530 		nfsrv_freenfslockfile(lfp);
1531 		ret = 1;
1532 	} else
1533 		ret = 0;
1534 	if (cansleep != 0)
1535 		NFSUNLOCKSTATE();
1536 	free(stp, M_NFSDSTATE);
1537 	NFSD_VNET(nfsstatsv1_p)->srvopens--;
1538 	nfsrv_openpluslock--;
1539 	return (ret);
1540 }
1541 
1542 /*
1543  * Frees a lockowner and all associated locks.
1544  */
1545 static void
1546 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1547     NFSPROC_T *p)
1548 {
1549 
1550 	LIST_REMOVE(stp, ls_hash);
1551 	LIST_REMOVE(stp, ls_list);
1552 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1553 	if (stp->ls_op)
1554 		nfsrvd_derefcache(stp->ls_op);
1555 	free(stp, M_NFSDSTATE);
1556 	NFSD_VNET(nfsstatsv1_p)->srvlockowners--;
1557 	nfsrv_openpluslock--;
1558 }
1559 
1560 /*
1561  * Free all the nfs locks on a lockowner.
1562  */
1563 static void
1564 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1565     NFSPROC_T *p)
1566 {
1567 	struct nfslock *lop, *nlop;
1568 	struct nfsrollback *rlp, *nrlp;
1569 	struct nfslockfile *lfp = NULL;
1570 	int gottvp = 0;
1571 	vnode_t tvp = NULL;
1572 	uint64_t first, end;
1573 
1574 	if (vp != NULL)
1575 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1576 	lop = LIST_FIRST(&stp->ls_lock);
1577 	while (lop != LIST_END(&stp->ls_lock)) {
1578 		nlop = LIST_NEXT(lop, lo_lckowner);
1579 		/*
1580 		 * Since all locks should be for the same file, lfp should
1581 		 * not change.
1582 		 */
1583 		if (lfp == NULL)
1584 			lfp = lop->lo_lfp;
1585 		else if (lfp != lop->lo_lfp)
1586 			panic("allnfslocks");
1587 		/*
1588 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1589 		 * from the file handle. This only occurs when called from
1590 		 * nfsrv_cleanclient().
1591 		 */
1592 		if (gottvp == 0) {
1593 			if (nfsrv_dolocallocks == 0)
1594 				tvp = NULL;
1595 			else if (vp == NULL && cansleep != 0) {
1596 				tvp = nfsvno_getvp(&lfp->lf_fh);
1597 				if (tvp != NULL)
1598 					NFSVOPUNLOCK(tvp);
1599 			} else
1600 				tvp = vp;
1601 			gottvp = 1;
1602 		}
1603 
1604 		if (tvp != NULL) {
1605 			if (cansleep == 0)
1606 				panic("allnfs2");
1607 			first = lop->lo_first;
1608 			end = lop->lo_end;
1609 			nfsrv_freenfslock(lop);
1610 			nfsrv_localunlock(tvp, lfp, first, end, p);
1611 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1612 			    nrlp)
1613 				free(rlp, M_NFSDROLLBACK);
1614 			LIST_INIT(&lfp->lf_rollback);
1615 		} else
1616 			nfsrv_freenfslock(lop);
1617 		lop = nlop;
1618 	}
1619 	if (vp == NULL && tvp != NULL)
1620 		vrele(tvp);
1621 }
1622 
1623 /*
1624  * Free an nfslock structure.
1625  */
1626 static void
1627 nfsrv_freenfslock(struct nfslock *lop)
1628 {
1629 
1630 	if (lop->lo_lckfile.le_prev != NULL) {
1631 		LIST_REMOVE(lop, lo_lckfile);
1632 		NFSD_VNET(nfsstatsv1_p)->srvlocks--;
1633 		nfsrv_openpluslock--;
1634 	}
1635 	LIST_REMOVE(lop, lo_lckowner);
1636 	free(lop, M_NFSDLOCK);
1637 }
1638 
1639 /*
1640  * This function frees an nfslockfile structure.
1641  */
1642 static void
1643 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1644 {
1645 
1646 	LIST_REMOVE(lfp, lf_hash);
1647 	free(lfp, M_NFSDLOCKFILE);
1648 }
1649 
1650 /*
1651  * This function looks up an nfsstate structure via stateid.
1652  */
1653 static int
1654 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1655     struct nfsstate **stpp)
1656 {
1657 	struct nfsstate *stp;
1658 	struct nfsstatehead *hp;
1659 	int error = 0;
1660 
1661 	*stpp = NULL;
1662 	hp = NFSSTATEHASH(clp, *stateidp);
1663 	LIST_FOREACH(stp, hp, ls_hash) {
1664 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1665 			NFSX_STATEIDOTHER))
1666 			break;
1667 	}
1668 
1669 	/*
1670 	 * If no state id in list, return NFSERR_BADSTATEID.
1671 	 */
1672 	if (stp == LIST_END(hp)) {
1673 		error = NFSERR_BADSTATEID;
1674 		goto out;
1675 	}
1676 	*stpp = stp;
1677 
1678 out:
1679 	NFSEXITCODE(error);
1680 	return (error);
1681 }
1682 
1683 /*
1684  * This function gets an nfsstate structure via owner string.
1685  */
1686 static void
1687 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1688     struct nfsstate **stpp)
1689 {
1690 	struct nfsstate *stp;
1691 
1692 	*stpp = NULL;
1693 	LIST_FOREACH(stp, hp, ls_list) {
1694 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1695 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1696 			*stpp = stp;
1697 			return;
1698 		}
1699 	}
1700 }
1701 
1702 /*
1703  * Lock control function called to update lock status.
1704  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1705  * that one isn't to be created and an NFSERR_xxx for other errors.
1706  * The structures new_stp and new_lop are passed in as pointers that should
1707  * be set to NULL if the structure is used and shouldn't be free'd.
1708  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1709  * never used and can safely be allocated on the stack. For all other
1710  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1711  * in case they are used.
1712  */
1713 int
1714 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1715     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1716     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1717     __unused struct nfsexstuff *exp,
1718     struct nfsrv_descript *nd, NFSPROC_T *p)
1719 {
1720 	struct nfslock *lop;
1721 	struct nfsstate *new_stp = *new_stpp;
1722 	struct nfslock *new_lop = *new_lopp;
1723 	struct nfsstate *tstp, *mystp, *nstp;
1724 	int specialid = 0;
1725 	struct nfslockfile *lfp;
1726 	struct nfslock *other_lop = NULL;
1727 	struct nfsstate *stp, *lckstp = NULL;
1728 	struct nfsclient *clp = NULL;
1729 	u_int32_t bits;
1730 	int error = 0, haslock = 0, ret, reterr;
1731 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1732 	fhandle_t nfh;
1733 	uint64_t first, end;
1734 	uint32_t lock_flags;
1735 
1736 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1737 		/*
1738 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1739 		 * let reads with all 1s go ahead.
1740 		 */
1741 		if (new_stp->ls_stateid.seqid == 0x0 &&
1742 		    new_stp->ls_stateid.other[0] == 0x0 &&
1743 		    new_stp->ls_stateid.other[1] == 0x0 &&
1744 		    new_stp->ls_stateid.other[2] == 0x0)
1745 			specialid = 1;
1746 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1747 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1748 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1749 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1750 			specialid = 2;
1751 	}
1752 
1753 	/*
1754 	 * Check for restart conditions (client and server).
1755 	 */
1756 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1757 	    &new_stp->ls_stateid, specialid);
1758 	if (error)
1759 		goto out;
1760 
1761 	/*
1762 	 * Check for state resource limit exceeded.
1763 	 */
1764 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1765 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1766 		error = NFSERR_RESOURCE;
1767 		goto out;
1768 	}
1769 
1770 	/*
1771 	 * For the lock case, get another nfslock structure,
1772 	 * just in case we need it.
1773 	 * Malloc now, before we start sifting through the linked lists,
1774 	 * in case we have to wait for memory.
1775 	 */
1776 tryagain:
1777 	if (new_stp->ls_flags & NFSLCK_LOCK)
1778 		other_lop = malloc(sizeof (struct nfslock),
1779 		    M_NFSDLOCK, M_WAITOK);
1780 	filestruct_locked = 0;
1781 	reterr = 0;
1782 	lfp = NULL;
1783 
1784 	/*
1785 	 * Get the lockfile structure for CFH now, so we can do a sanity
1786 	 * check against the stateid, before incrementing the seqid#, since
1787 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1788 	 * shouldn't be incremented for this case.
1789 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1790 	 * will be handled later.
1791 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1792 	 * lock the nfslockfile structure.
1793 	 */
1794 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1795 	NFSLOCKSTATE();
1796 	if (getlckret == 0) {
1797 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1798 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1799 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1800 			    &lfp, &nfh, 1);
1801 			if (getlckret == 0)
1802 				filestruct_locked = 1;
1803 		} else
1804 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1805 			    &lfp, &nfh, 0);
1806 	}
1807 	if (getlckret != 0 && getlckret != -1)
1808 		reterr = getlckret;
1809 
1810 	if (filestruct_locked != 0) {
1811 		LIST_INIT(&lfp->lf_rollback);
1812 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1813 			/*
1814 			 * For local locking, do the advisory locking now, so
1815 			 * that any conflict can be detected. A failure later
1816 			 * can be rolled back locally. If an error is returned,
1817 			 * struct nfslockfile has been unlocked and any local
1818 			 * locking rolled back.
1819 			 */
1820 			NFSUNLOCKSTATE();
1821 			if (vnode_unlocked == 0) {
1822 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1823 				vnode_unlocked = 1;
1824 				NFSVOPUNLOCK(vp);
1825 			}
1826 			reterr = nfsrv_locallock(vp, lfp,
1827 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1828 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1829 			NFSLOCKSTATE();
1830 		}
1831 	}
1832 
1833 	if (specialid == 0) {
1834 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1835 		/*
1836 		 * RFC 3530 does not list LockT as an op that renews a
1837 		 * lease, but the consensus seems to be that it is ok
1838 		 * for a server to do so.
1839 		 */
1840 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1841 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1842 
1843 		/*
1844 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1845 		 * error returns for LockT, just go ahead and test for a lock,
1846 		 * since there are no locks for this client, but other locks
1847 		 * can conflict. (ie. same client will always be false)
1848 		 */
1849 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1850 		    error = 0;
1851 		lckstp = new_stp;
1852 	    } else {
1853 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1854 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1855 	      if (error == 0)
1856 		/*
1857 		 * Look up the stateid
1858 		 */
1859 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1860 		  new_stp->ls_flags, &stp);
1861 	      /*
1862 	       * do some sanity checks for an unconfirmed open or a
1863 	       * stateid that refers to the wrong file, for an open stateid
1864 	       */
1865 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1866 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1867 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1868 		      /*
1869 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1870 		       * The only exception is using SETATTR with SIZE.
1871 		       * */
1872                     if ((new_stp->ls_flags &
1873                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1874 			     error = NFSERR_BADSTATEID;
1875 	      }
1876 
1877 		if (error == 0 &&
1878 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1879 		  getlckret == 0 && stp->ls_lfp != lfp)
1880 			error = NFSERR_BADSTATEID;
1881 
1882 	      /*
1883 	       * If the lockowner stateid doesn't refer to the same file,
1884 	       * I believe that is considered ok, since some clients will
1885 	       * only create a single lockowner and use that for all locks
1886 	       * on all files.
1887 	       * For now, log it as a diagnostic, instead of considering it
1888 	       * a BadStateid.
1889 	       */
1890 	      if (error == 0 && (stp->ls_flags &
1891 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1892 		  getlckret == 0 && stp->ls_lfp != lfp) {
1893 #ifdef DIAGNOSTIC
1894 		  printf("Got a lock statid for different file open\n");
1895 #endif
1896 		  /*
1897 		  error = NFSERR_BADSTATEID;
1898 		  */
1899 	      }
1900 
1901 	      if (error == 0) {
1902 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1903 			/*
1904 			 * If haslock set, we've already checked the seqid.
1905 			 */
1906 			if (!haslock) {
1907 			    if (stp->ls_flags & NFSLCK_OPEN)
1908 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1909 				    stp->ls_openowner, new_stp->ls_op);
1910 			    else
1911 				error = NFSERR_BADSTATEID;
1912 			}
1913 			if (!error)
1914 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1915 			if (lckstp) {
1916 			    /*
1917 			     * For NFSv4.1 and NFSv4.2 allow an
1918 			     * open_to_lock_owner when the lock_owner already
1919 			     * exists.  Just clear NFSLCK_OPENTOLOCK so that
1920 			     * a new lock_owner will not be created.
1921 			     * RFC7530 states that the error for NFSv4.0
1922 			     * is NFS4ERR_BAD_SEQID.
1923 			     */
1924 			    if ((nd->nd_flag & ND_NFSV41) != 0)
1925 				new_stp->ls_flags &= ~NFSLCK_OPENTOLOCK;
1926 			    else
1927 				error = NFSERR_BADSEQID;
1928 			} else
1929 			    lckstp = new_stp;
1930 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1931 			/*
1932 			 * If haslock set, ditto above.
1933 			 */
1934 			if (!haslock) {
1935 			    if (stp->ls_flags & NFSLCK_OPEN)
1936 				error = NFSERR_BADSTATEID;
1937 			    else
1938 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1939 				    stp, new_stp->ls_op);
1940 			}
1941 			lckstp = stp;
1942 		    } else {
1943 			lckstp = stp;
1944 		    }
1945 	      }
1946 	      /*
1947 	       * If the seqid part of the stateid isn't the same, return
1948 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1949 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1950 	       * nfsrv_returnoldstateid is set. (The consensus on the email
1951 	       * list was that most clients would prefer to not receive
1952 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1953 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1954 	       * allow for either server configuration.)
1955 	       */
1956 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1957 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1958 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1959 		    nfsrv_returnoldstateid)) ||
1960 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1961 		    new_stp->ls_stateid.seqid != 0)))
1962 		    error = NFSERR_OLDSTATEID;
1963 	    }
1964 	}
1965 
1966 	/*
1967 	 * Now we can check for grace.
1968 	 */
1969 	if (!error)
1970 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1971 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1972 		nfsrv_checkstable(clp))
1973 		error = NFSERR_NOGRACE;
1974 	/*
1975 	 * If we successfully Reclaimed state, note that.
1976 	 */
1977 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1978 		nfsrv_markstable(clp);
1979 
1980 	/*
1981 	 * At this point, either error == NFSERR_BADSTATEID or the
1982 	 * seqid# has been updated, so we can return any error.
1983 	 * If error == 0, there may be an error in:
1984 	 *    nd_repstat - Set by the calling function.
1985 	 *    reterr - Set above, if getting the nfslockfile structure
1986 	 *       or acquiring the local lock failed.
1987 	 *    (If both of these are set, nd_repstat should probably be
1988 	 *     returned, since that error was detected before this
1989 	 *     function call.)
1990 	 */
1991 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1992 		if (error == 0) {
1993 			if (nd->nd_repstat != 0)
1994 				error = nd->nd_repstat;
1995 			else
1996 				error = reterr;
1997 		}
1998 		if (filestruct_locked != 0) {
1999 			/* Roll back local locks. */
2000 			NFSUNLOCKSTATE();
2001 			if (vnode_unlocked == 0) {
2002 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
2003 				vnode_unlocked = 1;
2004 				NFSVOPUNLOCK(vp);
2005 			}
2006 			nfsrv_locallock_rollback(vp, lfp, p);
2007 			NFSLOCKSTATE();
2008 			nfsrv_unlocklf(lfp);
2009 		}
2010 		NFSUNLOCKSTATE();
2011 		goto out;
2012 	}
2013 
2014 	/*
2015 	 * Check the nfsrv_getlockfile return.
2016 	 * Returned -1 if no structure found.
2017 	 */
2018 	if (getlckret == -1) {
2019 		error = NFSERR_EXPIRED;
2020 		/*
2021 		 * Called from lockt, so no lock is OK.
2022 		 */
2023 		if (new_stp->ls_flags & NFSLCK_TEST) {
2024 			error = 0;
2025 		} else if (new_stp->ls_flags &
2026 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
2027 			/*
2028 			 * Called to check for a lock, OK if the stateid is all
2029 			 * 1s or all 0s, but there should be an nfsstate
2030 			 * otherwise.
2031 			 * (ie. If there is no open, I'll assume no share
2032 			 *  deny bits.)
2033 			 */
2034 			if (specialid)
2035 				error = 0;
2036 			else
2037 				error = NFSERR_BADSTATEID;
2038 		}
2039 		NFSUNLOCKSTATE();
2040 		goto out;
2041 	}
2042 
2043 	/*
2044 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
2045 	 * For NFSLCK_CHECK, allow a read if write access is granted,
2046 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
2047 	 * which implies a conflicting deny can't exist.
2048 	 */
2049 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
2050 	    /*
2051 	     * Four kinds of state id:
2052 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
2053 	     * - stateid for an open
2054 	     * - stateid for a delegation
2055 	     * - stateid for a lock owner
2056 	     */
2057 	    if (!specialid) {
2058 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2059 		    delegation = 1;
2060 		    mystp = stp;
2061 		    nfsrv_delaydelegtimeout(stp);
2062 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
2063 		    mystp = stp;
2064 		} else {
2065 		    mystp = stp->ls_openstp;
2066 		}
2067 		/*
2068 		 * If locking or checking, require correct access
2069 		 * bit set.
2070 		 */
2071 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
2072 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
2073 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
2074 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
2075 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
2076 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
2077 		     nfsrv_allowreadforwriteopen == 0) ||
2078 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
2079 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
2080 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
2081 			if (filestruct_locked != 0) {
2082 				/* Roll back local locks. */
2083 				NFSUNLOCKSTATE();
2084 				if (vnode_unlocked == 0) {
2085 					ASSERT_VOP_ELOCKED(vp,
2086 					    "nfsrv_lockctrl3");
2087 					vnode_unlocked = 1;
2088 					NFSVOPUNLOCK(vp);
2089 				}
2090 				nfsrv_locallock_rollback(vp, lfp, p);
2091 				NFSLOCKSTATE();
2092 				nfsrv_unlocklf(lfp);
2093 			}
2094 			NFSUNLOCKSTATE();
2095 			error = NFSERR_OPENMODE;
2096 			goto out;
2097 		}
2098 	    } else
2099 		mystp = NULL;
2100 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
2101 		/*
2102 		 * Check for a conflicting deny bit.
2103 		 */
2104 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
2105 		    if (tstp != mystp) {
2106 			bits = tstp->ls_flags;
2107 			bits >>= NFSLCK_SHIFT;
2108 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
2109 			    KASSERT(vnode_unlocked == 0,
2110 				("nfsrv_lockctrl: vnode unlocked1"));
2111 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
2112 				vp, p);
2113 			    if (ret == 1) {
2114 				/*
2115 				* nfsrv_clientconflict unlocks state
2116 				 * when it returns non-zero.
2117 				 */
2118 				lckstp = NULL;
2119 				goto tryagain;
2120 			    }
2121 			    if (ret == 0)
2122 				NFSUNLOCKSTATE();
2123 			    if (ret == 2)
2124 				error = NFSERR_PERM;
2125 			    else
2126 				error = NFSERR_OPENMODE;
2127 			    goto out;
2128 			}
2129 		    }
2130 		}
2131 
2132 		/* We're outta here */
2133 		NFSUNLOCKSTATE();
2134 		goto out;
2135 	    }
2136 	}
2137 
2138 	/*
2139 	 * For setattr, just get rid of all the Delegations for other clients.
2140 	 */
2141 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
2142 		KASSERT(vnode_unlocked == 0,
2143 		    ("nfsrv_lockctrl: vnode unlocked2"));
2144 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2145 		if (ret) {
2146 			/*
2147 			 * nfsrv_cleandeleg() unlocks state when it
2148 			 * returns non-zero.
2149 			 */
2150 			if (ret == -1) {
2151 				lckstp = NULL;
2152 				goto tryagain;
2153 			}
2154 			error = ret;
2155 			goto out;
2156 		}
2157 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2158 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2159 		     LIST_EMPTY(&lfp->lf_deleg))) {
2160 			NFSUNLOCKSTATE();
2161 			goto out;
2162 		}
2163 	}
2164 
2165 	/*
2166 	 * Check for a conflicting delegation. If one is found, call
2167 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2168 	 * been set yet, it will get the lock. Otherwise, it will recall
2169 	 * the delegation. Then, we try try again...
2170 	 * I currently believe the conflict algorithm to be:
2171 	 * For Lock Ops (Lock/LockT/LockU)
2172 	 * - there is a conflict iff a different client has a write delegation
2173 	 * For Reading (Read Op)
2174 	 * - there is a conflict iff a different client has a write delegation
2175 	 *   (the specialids are always a different client)
2176 	 * For Writing (Write/Setattr of size)
2177 	 * - there is a conflict if a different client has any delegation
2178 	 * - there is a conflict if the same client has a read delegation
2179 	 *   (I don't understand why this isn't allowed, but that seems to be
2180 	 *    the current consensus?)
2181 	 */
2182 	tstp = LIST_FIRST(&lfp->lf_deleg);
2183 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2184 	    nstp = LIST_NEXT(tstp, ls_file);
2185 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2186 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2187 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2188 		  clp != tstp->ls_clp &&
2189 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2190 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2191 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2192 		  (clp != tstp->ls_clp ||
2193 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2194 		ret = 0;
2195 		if (filestruct_locked != 0) {
2196 			/* Roll back local locks. */
2197 			NFSUNLOCKSTATE();
2198 			if (vnode_unlocked == 0) {
2199 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2200 				NFSVOPUNLOCK(vp);
2201 			}
2202 			nfsrv_locallock_rollback(vp, lfp, p);
2203 			NFSLOCKSTATE();
2204 			nfsrv_unlocklf(lfp);
2205 			NFSUNLOCKSTATE();
2206 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2207 			vnode_unlocked = 0;
2208 			if (VN_IS_DOOMED(vp))
2209 				ret = NFSERR_SERVERFAULT;
2210 			NFSLOCKSTATE();
2211 		}
2212 		if (ret == 0)
2213 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2214 		if (ret) {
2215 		    /*
2216 		     * nfsrv_delegconflict unlocks state when it
2217 		     * returns non-zero, which it always does.
2218 		     */
2219 		    if (other_lop) {
2220 			free(other_lop, M_NFSDLOCK);
2221 			other_lop = NULL;
2222 		    }
2223 		    if (ret == -1) {
2224 			lckstp = NULL;
2225 			goto tryagain;
2226 		    }
2227 		    error = ret;
2228 		    goto out;
2229 		}
2230 		/* Never gets here. */
2231 	    }
2232 	    tstp = nstp;
2233 	}
2234 
2235 	/*
2236 	 * Handle the unlock case by calling nfsrv_updatelock().
2237 	 * (Should I have done some access checking above for unlock? For now,
2238 	 *  just let it happen.)
2239 	 */
2240 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2241 		first = new_lop->lo_first;
2242 		end = new_lop->lo_end;
2243 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2244 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2245 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2246 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2247 		stateidp->other[0] = stp->ls_stateid.other[0];
2248 		stateidp->other[1] = stp->ls_stateid.other[1];
2249 		stateidp->other[2] = stp->ls_stateid.other[2];
2250 		if (filestruct_locked != 0) {
2251 			NFSUNLOCKSTATE();
2252 			if (vnode_unlocked == 0) {
2253 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2254 				vnode_unlocked = 1;
2255 				NFSVOPUNLOCK(vp);
2256 			}
2257 			/* Update the local locks. */
2258 			nfsrv_localunlock(vp, lfp, first, end, p);
2259 			NFSLOCKSTATE();
2260 			nfsrv_unlocklf(lfp);
2261 		}
2262 		NFSUNLOCKSTATE();
2263 		goto out;
2264 	}
2265 
2266 	/*
2267 	 * Search for a conflicting lock. A lock conflicts if:
2268 	 * - the lock range overlaps and
2269 	 * - at least one lock is a write lock and
2270 	 * - it is not owned by the same lock owner
2271 	 */
2272 	if (!delegation) {
2273 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2274 	    if (new_lop->lo_end > lop->lo_first &&
2275 		new_lop->lo_first < lop->lo_end &&
2276 		(new_lop->lo_flags == NFSLCK_WRITE ||
2277 		 lop->lo_flags == NFSLCK_WRITE) &&
2278 		lckstp != lop->lo_stp &&
2279 		(clp != lop->lo_stp->ls_clp ||
2280 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2281 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2282 		    lckstp->ls_ownerlen))) {
2283 		if (other_lop) {
2284 		    free(other_lop, M_NFSDLOCK);
2285 		    other_lop = NULL;
2286 		}
2287 		if (vnode_unlocked != 0)
2288 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2289 			NULL, p);
2290 		else
2291 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2292 			vp, p);
2293 		if (ret == 1) {
2294 		    if (filestruct_locked != 0) {
2295 			if (vnode_unlocked == 0) {
2296 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2297 				NFSVOPUNLOCK(vp);
2298 			}
2299 			/* Roll back local locks. */
2300 			nfsrv_locallock_rollback(vp, lfp, p);
2301 			NFSLOCKSTATE();
2302 			nfsrv_unlocklf(lfp);
2303 			NFSUNLOCKSTATE();
2304 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2305 			vnode_unlocked = 0;
2306 			if (VN_IS_DOOMED(vp)) {
2307 				error = NFSERR_SERVERFAULT;
2308 				goto out;
2309 			}
2310 		    }
2311 		    /*
2312 		     * nfsrv_clientconflict() unlocks state when it
2313 		     * returns non-zero.
2314 		     */
2315 		    lckstp = NULL;
2316 		    goto tryagain;
2317 		}
2318 		/*
2319 		 * Found a conflicting lock, so record the conflict and
2320 		 * return the error.
2321 		 */
2322 		if (cfp != NULL && ret == 0) {
2323 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2324 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2325 		    cfp->cl_first = lop->lo_first;
2326 		    cfp->cl_end = lop->lo_end;
2327 		    cfp->cl_flags = lop->lo_flags;
2328 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2329 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2330 			cfp->cl_ownerlen);
2331 		}
2332 		if (ret == 2)
2333 		    error = NFSERR_PERM;
2334 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2335 		    error = NFSERR_RECLAIMCONFLICT;
2336 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2337 		    error = NFSERR_LOCKED;
2338 		else
2339 		    error = NFSERR_DENIED;
2340 		if (filestruct_locked != 0 && ret == 0) {
2341 			/* Roll back local locks. */
2342 			NFSUNLOCKSTATE();
2343 			if (vnode_unlocked == 0) {
2344 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2345 				vnode_unlocked = 1;
2346 				NFSVOPUNLOCK(vp);
2347 			}
2348 			nfsrv_locallock_rollback(vp, lfp, p);
2349 			NFSLOCKSTATE();
2350 			nfsrv_unlocklf(lfp);
2351 		}
2352 		if (ret == 0)
2353 			NFSUNLOCKSTATE();
2354 		goto out;
2355 	    }
2356 	  }
2357 	}
2358 
2359 	/*
2360 	 * We only get here if there was no lock that conflicted.
2361 	 */
2362 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2363 		NFSUNLOCKSTATE();
2364 		goto out;
2365 	}
2366 
2367 	/*
2368 	 * We only get here when we are creating or modifying a lock.
2369 	 * There are two variants:
2370 	 * - exist_lock_owner where lock_owner exists
2371 	 * - open_to_lock_owner with new lock_owner
2372 	 */
2373 	first = new_lop->lo_first;
2374 	end = new_lop->lo_end;
2375 	lock_flags = new_lop->lo_flags;
2376 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2377 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2378 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2379 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2380 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2381 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2382 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2383 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2384 	} else {
2385 		/*
2386 		 * The new open_to_lock_owner case.
2387 		 * Link the new nfsstate into the lists.
2388 		 */
2389 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2390 		nfsrvd_refcache(new_stp->ls_op);
2391 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2392 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2393 		    clp->lc_clientid.lval[0];
2394 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2395 		    clp->lc_clientid.lval[1];
2396 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2397 		    nfsrv_nextstateindex(clp);
2398 		new_stp->ls_clp = clp;
2399 		LIST_INIT(&new_stp->ls_lock);
2400 		new_stp->ls_openstp = stp;
2401 		new_stp->ls_lfp = lfp;
2402 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2403 		    lfp);
2404 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2405 		    new_stp, ls_hash);
2406 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2407 		*new_lopp = NULL;
2408 		*new_stpp = NULL;
2409 		NFSD_VNET(nfsstatsv1_p)->srvlockowners++;
2410 		nfsrv_openpluslock++;
2411 	}
2412 	if (filestruct_locked != 0) {
2413 		NFSUNLOCKSTATE();
2414 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2415 		NFSLOCKSTATE();
2416 		nfsrv_unlocklf(lfp);
2417 	}
2418 	NFSUNLOCKSTATE();
2419 
2420 out:
2421 	if (haslock) {
2422 		NFSLOCKV4ROOTMUTEX();
2423 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2424 		NFSUNLOCKV4ROOTMUTEX();
2425 	}
2426 	if (vnode_unlocked != 0) {
2427 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2428 		if (error == 0 && VN_IS_DOOMED(vp))
2429 			error = NFSERR_SERVERFAULT;
2430 	}
2431 	if (other_lop)
2432 		free(other_lop, M_NFSDLOCK);
2433 	NFSEXITCODE2(error, nd);
2434 	return (error);
2435 }
2436 
2437 /*
2438  * Check for state errors for Open.
2439  * repstat is passed back out as an error if more critical errors
2440  * are not detected.
2441  */
2442 int
2443 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2444     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2445     NFSPROC_T *p, int repstat)
2446 {
2447 	struct nfsstate *stp, *nstp;
2448 	struct nfsclient *clp;
2449 	struct nfsstate *ownerstp;
2450 	struct nfslockfile *lfp, *new_lfp;
2451 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2452 
2453 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2454 		readonly = 1;
2455 	/*
2456 	 * Check for restart conditions (client and server).
2457 	 */
2458 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2459 		&new_stp->ls_stateid, 0);
2460 	if (error)
2461 		goto out;
2462 
2463 	/*
2464 	 * Check for state resource limit exceeded.
2465 	 * Technically this should be SMP protected, but the worst
2466 	 * case error is "out by one or two" on the count when it
2467 	 * returns NFSERR_RESOURCE and the limit is just a rather
2468 	 * arbitrary high water mark, so no harm is done.
2469 	 */
2470 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2471 		error = NFSERR_RESOURCE;
2472 		goto out;
2473 	}
2474 
2475 tryagain:
2476 	new_lfp = malloc(sizeof (struct nfslockfile),
2477 	    M_NFSDLOCKFILE, M_WAITOK);
2478 	if (vp)
2479 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2480 		    NULL, p);
2481 	NFSLOCKSTATE();
2482 	/*
2483 	 * Get the nfsclient structure.
2484 	 */
2485 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2486 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2487 
2488 	/*
2489 	 * Look up the open owner. See if it needs confirmation and
2490 	 * check the seq#, as required.
2491 	 */
2492 	if (!error)
2493 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2494 
2495 	if (!error && ownerstp) {
2496 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2497 		    new_stp->ls_op);
2498 		/*
2499 		 * If the OpenOwner hasn't been confirmed, assume the
2500 		 * old one was a replay and this one is ok.
2501 		 * See: RFC3530 Sec. 14.2.18.
2502 		 */
2503 		if (error == NFSERR_BADSEQID &&
2504 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2505 			error = 0;
2506 	}
2507 
2508 	/*
2509 	 * Check for grace.
2510 	 */
2511 	if (!error)
2512 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2513 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2514 		nfsrv_checkstable(clp))
2515 		error = NFSERR_NOGRACE;
2516 
2517 	/*
2518 	 * If none of the above errors occurred, let repstat be
2519 	 * returned.
2520 	 */
2521 	if (repstat && !error)
2522 		error = repstat;
2523 	if (error) {
2524 		NFSUNLOCKSTATE();
2525 		if (haslock) {
2526 			NFSLOCKV4ROOTMUTEX();
2527 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2528 			NFSUNLOCKV4ROOTMUTEX();
2529 		}
2530 		free(new_lfp, M_NFSDLOCKFILE);
2531 		goto out;
2532 	}
2533 
2534 	/*
2535 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2536 	 * (This always happens on the first pass, so haslock must be 0.)
2537 	 */
2538 	if (vp == NULL) {
2539 		NFSUNLOCKSTATE();
2540 		free(new_lfp, M_NFSDLOCKFILE);
2541 		goto out;
2542 	}
2543 
2544 	/*
2545 	 * Get the structure for the underlying file.
2546 	 */
2547 	if (getfhret)
2548 		error = getfhret;
2549 	else
2550 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2551 		    NULL, 0);
2552 	if (new_lfp)
2553 		free(new_lfp, M_NFSDLOCKFILE);
2554 	if (error) {
2555 		NFSUNLOCKSTATE();
2556 		if (haslock) {
2557 			NFSLOCKV4ROOTMUTEX();
2558 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2559 			NFSUNLOCKV4ROOTMUTEX();
2560 		}
2561 		goto out;
2562 	}
2563 
2564 	/*
2565 	 * Search for a conflicting open/share.
2566 	 */
2567 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2568 	    /*
2569 	     * For Delegate_Cur, search for the matching Delegation,
2570 	     * which indicates no conflict.
2571 	     * For NFSv4.1/4.2 Claim_Deleg_Cur_FH only provides
2572 	     * the clientid, which is the first two "other" elements
2573 	     * for the stateid.  This should be sufficient, since there
2574 	     * is only one delegation per client and file.
2575 	     * An old delegation should have been recovered by the
2576 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2577 	     * it match and return NFSERR_EXPIRED. Should I let it
2578 	     * match?
2579 	     */
2580 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2581 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2582 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2583 		    stateidp->seqid == 0) ||
2584 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2585 		    stateidp->other[0] == stp->ls_stateid.other[0] &&
2586 		    stateidp->other[1] == stp->ls_stateid.other[1])
2587 			break;
2588 	    }
2589 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2590 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2591 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2592 		NFSUNLOCKSTATE();
2593 		if (haslock) {
2594 			NFSLOCKV4ROOTMUTEX();
2595 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2596 			NFSUNLOCKV4ROOTMUTEX();
2597 		}
2598 		error = NFSERR_EXPIRED;
2599 		goto out;
2600 	    }
2601 	}
2602 
2603 	/*
2604 	 * Check for access/deny bit conflicts. I check for the same
2605 	 * owner as well, in case the client didn't bother.
2606 	 */
2607 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2608 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2609 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2610 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2611 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2612 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2613 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2614 			if (ret == 1) {
2615 				/*
2616 				 * nfsrv_clientconflict() unlocks
2617 				 * state when it returns non-zero.
2618 				 */
2619 				goto tryagain;
2620 			}
2621 			if (ret == 2)
2622 				error = NFSERR_PERM;
2623 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2624 				error = NFSERR_RECLAIMCONFLICT;
2625 			else
2626 				error = NFSERR_SHAREDENIED;
2627 			if (ret == 0)
2628 				NFSUNLOCKSTATE();
2629 			if (haslock) {
2630 				NFSLOCKV4ROOTMUTEX();
2631 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2632 				NFSUNLOCKV4ROOTMUTEX();
2633 			}
2634 			goto out;
2635 		}
2636 	}
2637 
2638 	/*
2639 	 * Check for a conflicting delegation. If one is found, call
2640 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2641 	 * been set yet, it will get the lock. Otherwise, it will recall
2642 	 * the delegation. Then, we try try again...
2643 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2644 	 *  isn't a conflict.)
2645 	 * I currently believe the conflict algorithm to be:
2646 	 * For Open with Read Access and Deny None
2647 	 * - there is a conflict iff a different client has a write delegation
2648 	 * For Open with other Write Access or any Deny except None
2649 	 * - there is a conflict if a different client has any delegation
2650 	 * - there is a conflict if the same client has a read delegation
2651 	 *   (The current consensus is that this last case should be
2652 	 *    considered a conflict since the client with a read delegation
2653 	 *    could have done an Open with ReadAccess and WriteDeny
2654 	 *    locally and then not have checked for the WriteDeny.)
2655 	 * Don't check for a Reclaim, since that will be dealt with
2656 	 * by nfsrv_openctrl().
2657 	 */
2658 	if (!(new_stp->ls_flags &
2659 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2660 	    stp = LIST_FIRST(&lfp->lf_deleg);
2661 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2662 		nstp = LIST_NEXT(stp, ls_file);
2663 		if ((readonly && stp->ls_clp != clp &&
2664 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2665 		    (!readonly && (stp->ls_clp != clp ||
2666 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2667 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2668 			if (ret) {
2669 			    /*
2670 			     * nfsrv_delegconflict() unlocks state
2671 			     * when it returns non-zero.
2672 			     */
2673 			    if (ret == -1)
2674 				goto tryagain;
2675 			    error = ret;
2676 			    goto out;
2677 			}
2678 		}
2679 		stp = nstp;
2680 	    }
2681 	}
2682 	NFSUNLOCKSTATE();
2683 	if (haslock) {
2684 		NFSLOCKV4ROOTMUTEX();
2685 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2686 		NFSUNLOCKV4ROOTMUTEX();
2687 	}
2688 
2689 out:
2690 	NFSEXITCODE2(error, nd);
2691 	return (error);
2692 }
2693 
2694 /*
2695  * Open control function to create/update open state for an open.
2696  */
2697 int
2698 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2699     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2700     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2701     NFSPROC_T *p, u_quad_t filerev)
2702 {
2703 	struct nfsstate *new_stp = *new_stpp;
2704 	struct nfsstate *stp, *nstp;
2705 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2706 	struct nfslockfile *lfp, *new_lfp;
2707 	struct nfsclient *clp;
2708 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2709 	int readonly = 0, cbret = 1, getfhret = 0;
2710 	int gotstate = 0, len = 0;
2711 	u_char *clidp = NULL;
2712 
2713 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2714 		readonly = 1;
2715 	/*
2716 	 * Check for restart conditions (client and server).
2717 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2718 	 * If an error does show up, return NFSERR_EXPIRED, since the
2719 	 * the seqid# has already been incremented.
2720 	 */
2721 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2722 	    &new_stp->ls_stateid, 0);
2723 	if (error) {
2724 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2725 		    error);
2726 		error = NFSERR_EXPIRED;
2727 		goto out;
2728 	}
2729 
2730 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2731 tryagain:
2732 	new_lfp = malloc(sizeof (struct nfslockfile),
2733 	    M_NFSDLOCKFILE, M_WAITOK);
2734 	new_open = malloc(sizeof (struct nfsstate),
2735 	    M_NFSDSTATE, M_WAITOK);
2736 	new_deleg = malloc(sizeof (struct nfsstate),
2737 	    M_NFSDSTATE, M_WAITOK);
2738 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2739 	    NULL, p);
2740 	NFSLOCKSTATE();
2741 	/*
2742 	 * Get the client structure. Since the linked lists could be changed
2743 	 * by other nfsd processes if this process does a tsleep(), one of
2744 	 * two things must be done.
2745 	 * 1 - don't tsleep()
2746 	 * or
2747 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2748 	 *     before using the lists, since this lock stops the other
2749 	 *     nfsd. This should only be used for rare cases, since it
2750 	 *     essentially single threads the nfsd.
2751 	 *     At this time, it is only done for cases where the stable
2752 	 *     storage file must be written prior to completion of state
2753 	 *     expiration.
2754 	 */
2755 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2756 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2757 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2758 	    clp->lc_program) {
2759 		/*
2760 		 * This happens on the first open for a client
2761 		 * that supports callbacks.
2762 		 */
2763 		NFSUNLOCKSTATE();
2764 		/*
2765 		 * Although nfsrv_docallback() will sleep, clp won't
2766 		 * go away, since they are only removed when the
2767 		 * nfsv4_lock() has blocked the nfsd threads. The
2768 		 * fields in clp can change, but having multiple
2769 		 * threads do this Null callback RPC should be
2770 		 * harmless.
2771 		 */
2772 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2773 		    NULL, 0, NULL, NULL, NULL, 0, p);
2774 		NFSLOCKSTATE();
2775 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2776 		if (!cbret)
2777 			clp->lc_flags |= LCL_CALLBACKSON;
2778 	}
2779 
2780 	/*
2781 	 * Look up the open owner. See if it needs confirmation and
2782 	 * check the seq#, as required.
2783 	 */
2784 	if (!error)
2785 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2786 
2787 	if (error) {
2788 		NFSUNLOCKSTATE();
2789 		printf("Nfsd: openctrl unexpected state err=%d\n",
2790 			error);
2791 		free(new_lfp, M_NFSDLOCKFILE);
2792 		free(new_open, M_NFSDSTATE);
2793 		free(new_deleg, M_NFSDSTATE);
2794 		if (haslock) {
2795 			NFSLOCKV4ROOTMUTEX();
2796 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2797 			NFSUNLOCKV4ROOTMUTEX();
2798 		}
2799 		error = NFSERR_EXPIRED;
2800 		goto out;
2801 	}
2802 
2803 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2804 		nfsrv_markstable(clp);
2805 
2806 	/*
2807 	 * Get the structure for the underlying file.
2808 	 */
2809 	if (getfhret)
2810 		error = getfhret;
2811 	else
2812 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2813 		    NULL, 0);
2814 	if (new_lfp)
2815 		free(new_lfp, M_NFSDLOCKFILE);
2816 	if (error) {
2817 		NFSUNLOCKSTATE();
2818 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2819 		    error);
2820 		free(new_open, M_NFSDSTATE);
2821 		free(new_deleg, M_NFSDSTATE);
2822 		if (haslock) {
2823 			NFSLOCKV4ROOTMUTEX();
2824 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2825 			NFSUNLOCKV4ROOTMUTEX();
2826 		}
2827 		goto out;
2828 	}
2829 
2830 	/*
2831 	 * Search for a conflicting open/share.
2832 	 */
2833 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2834 	    /*
2835 	     * For Delegate_Cur, search for the matching Delegation,
2836 	     * which indicates no conflict.
2837 	     * For NFSv4.1/4.2 Claim_Deleg_Cur_FH only provides
2838 	     * the clientid, which is the first two "other" elements
2839 	     * for the stateid.  This should be sufficient, since there
2840 	     * is only one delegation per client and file.
2841 	     * An old delegation should have been recovered by the
2842 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2843 	     * it match and return NFSERR_EXPIRED. Should I let it
2844 	     * match?
2845 	     */
2846 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2847 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2848 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2849 		    stateidp->seqid == 0) ||
2850 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2851 		    stateidp->other[0] == stp->ls_stateid.other[0] &&
2852 		    stateidp->other[1] == stp->ls_stateid.other[1])
2853 			break;
2854 	    }
2855 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2856 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2857 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2858 		NFSUNLOCKSTATE();
2859 		printf("Nfsd openctrl unexpected expiry\n");
2860 		free(new_open, M_NFSDSTATE);
2861 		free(new_deleg, M_NFSDSTATE);
2862 		if (haslock) {
2863 			NFSLOCKV4ROOTMUTEX();
2864 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2865 			NFSUNLOCKV4ROOTMUTEX();
2866 		}
2867 		error = NFSERR_EXPIRED;
2868 		goto out;
2869 	    }
2870 
2871 	    /*
2872 	     * Don't issue a Delegation, since one already exists and
2873 	     * delay delegation timeout, as required.
2874 	     */
2875 	    delegate = 0;
2876 	    nfsrv_delaydelegtimeout(stp);
2877 	}
2878 
2879 	/*
2880 	 * Check for access/deny bit conflicts. I also check for the
2881 	 * same owner, since the client might not have bothered to check.
2882 	 * Also, note an open for the same file and owner, if found,
2883 	 * which is all we do here for Delegate_Cur, since conflict
2884 	 * checking is already done.
2885 	 */
2886 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2887 		if (ownerstp && stp->ls_openowner == ownerstp)
2888 			openstp = stp;
2889 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2890 		    /*
2891 		     * If another client has the file open, the only
2892 		     * delegation that can be issued is a Read delegation
2893 		     * and only if it is a Read open with Deny none.
2894 		     */
2895 		    if (clp != stp->ls_clp) {
2896 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2897 			    NFSLCK_READACCESS)
2898 			    writedeleg = 0;
2899 			else
2900 			    delegate = 0;
2901 		    }
2902 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2903 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2904 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2905 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2906 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2907 			if (ret == 1) {
2908 				/*
2909 				 * nfsrv_clientconflict() unlocks state
2910 				 * when it returns non-zero.
2911 				 */
2912 				free(new_open, M_NFSDSTATE);
2913 				free(new_deleg, M_NFSDSTATE);
2914 				openstp = NULL;
2915 				goto tryagain;
2916 			}
2917 			if (ret == 2)
2918 				error = NFSERR_PERM;
2919 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2920 				error = NFSERR_RECLAIMCONFLICT;
2921 			else
2922 				error = NFSERR_SHAREDENIED;
2923 			if (ret == 0)
2924 				NFSUNLOCKSTATE();
2925 			if (haslock) {
2926 				NFSLOCKV4ROOTMUTEX();
2927 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2928 				NFSUNLOCKV4ROOTMUTEX();
2929 			}
2930 			free(new_open, M_NFSDSTATE);
2931 			free(new_deleg, M_NFSDSTATE);
2932 			printf("nfsd openctrl unexpected client cnfl\n");
2933 			goto out;
2934 		    }
2935 		}
2936 	}
2937 
2938 	/*
2939 	 * Check for a conflicting delegation. If one is found, call
2940 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2941 	 * been set yet, it will get the lock. Otherwise, it will recall
2942 	 * the delegation. Then, we try try again...
2943 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2944 	 *  isn't a conflict.)
2945 	 * I currently believe the conflict algorithm to be:
2946 	 * For Open with Read Access and Deny None
2947 	 * - there is a conflict iff a different client has a write delegation
2948 	 * For Open with other Write Access or any Deny except None
2949 	 * - there is a conflict if a different client has any delegation
2950 	 * - there is a conflict if the same client has a read delegation
2951 	 *   (The current consensus is that this last case should be
2952 	 *    considered a conflict since the client with a read delegation
2953 	 *    could have done an Open with ReadAccess and WriteDeny
2954 	 *    locally and then not have checked for the WriteDeny.)
2955 	 */
2956 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2957 	    stp = LIST_FIRST(&lfp->lf_deleg);
2958 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2959 		nstp = LIST_NEXT(stp, ls_file);
2960 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2961 			writedeleg = 0;
2962 		else
2963 			delegate = 0;
2964 		if ((readonly && stp->ls_clp != clp &&
2965 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2966 		    (!readonly && (stp->ls_clp != clp ||
2967 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2968 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2969 			delegate = 2;
2970 		    } else {
2971 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2972 			if (ret) {
2973 			    /*
2974 			     * nfsrv_delegconflict() unlocks state
2975 			     * when it returns non-zero.
2976 			     */
2977 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2978 			    free(new_open, M_NFSDSTATE);
2979 			    free(new_deleg, M_NFSDSTATE);
2980 			    if (ret == -1) {
2981 				openstp = NULL;
2982 				goto tryagain;
2983 			    }
2984 			    error = ret;
2985 			    goto out;
2986 			}
2987 		    }
2988 		}
2989 		stp = nstp;
2990 	    }
2991 	}
2992 
2993 	/*
2994 	 * We only get here if there was no open that conflicted.
2995 	 * If an open for the owner exists, or in the access/deny bits.
2996 	 * Otherwise it is a new open. If the open_owner hasn't been
2997 	 * confirmed, replace the open with the new one needing confirmation,
2998 	 * otherwise add the open.
2999 	 */
3000 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
3001 	    /*
3002 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
3003 	     * a match. If found, just move the old delegation to the current
3004 	     * delegation list and issue open. If not found, return
3005 	     * NFSERR_EXPIRED.
3006 	     */
3007 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
3008 		if (stp->ls_lfp == lfp) {
3009 		    /* Found it */
3010 		    if (stp->ls_clp != clp)
3011 			panic("olddeleg clp");
3012 		    LIST_REMOVE(stp, ls_list);
3013 		    LIST_REMOVE(stp, ls_hash);
3014 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
3015 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
3016 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
3017 			clp->lc_clientid.lval[0];
3018 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
3019 			clp->lc_clientid.lval[1];
3020 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
3021 			nfsrv_nextstateindex(clp);
3022 		    stp->ls_compref = nd->nd_compref;
3023 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
3024 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3025 			stp->ls_stateid), stp, ls_hash);
3026 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3027 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3028 		    else
3029 			*rflagsp |= NFSV4OPEN_READDELEGATE;
3030 		    clp->lc_delegtime = NFSD_MONOSEC +
3031 			nfsrv_lease + NFSRV_LEASEDELTA;
3032 
3033 		    /*
3034 		     * Now, do the associated open.
3035 		     */
3036 		    new_open->ls_stateid.seqid = 1;
3037 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3038 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3039 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3040 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
3041 			NFSLCK_OPEN;
3042 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3043 			new_open->ls_flags |= (NFSLCK_READACCESS |
3044 			    NFSLCK_WRITEACCESS);
3045 		    else
3046 			new_open->ls_flags |= NFSLCK_READACCESS;
3047 		    new_open->ls_uid = new_stp->ls_uid;
3048 		    new_open->ls_lfp = lfp;
3049 		    new_open->ls_clp = clp;
3050 		    LIST_INIT(&new_open->ls_open);
3051 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3052 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3053 			new_open, ls_hash);
3054 		    /*
3055 		     * and handle the open owner
3056 		     */
3057 		    if (ownerstp) {
3058 			new_open->ls_openowner = ownerstp;
3059 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
3060 		    } else {
3061 			new_open->ls_openowner = new_stp;
3062 			new_stp->ls_flags = 0;
3063 			nfsrvd_refcache(new_stp->ls_op);
3064 			new_stp->ls_noopens = 0;
3065 			LIST_INIT(&new_stp->ls_open);
3066 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3067 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3068 			*new_stpp = NULL;
3069 			NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3070 			nfsrv_openpluslock++;
3071 		    }
3072 		    openstp = new_open;
3073 		    new_open = NULL;
3074 		    NFSD_VNET(nfsstatsv1_p)->srvopens++;
3075 		    nfsrv_openpluslock++;
3076 		    break;
3077 		}
3078 	    }
3079 	    if (stp == LIST_END(&clp->lc_olddeleg))
3080 		error = NFSERR_EXPIRED;
3081 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
3082 	    /*
3083 	     * Scan to see that no delegation for this client and file
3084 	     * doesn't already exist.
3085 	     * There also shouldn't yet be an Open for this file and
3086 	     * openowner.
3087 	     */
3088 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
3089 		if (stp->ls_clp == clp)
3090 		    break;
3091 	    }
3092 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
3093 		/*
3094 		 * This is the Claim_Previous case with a delegation
3095 		 * type != Delegate_None.
3096 		 */
3097 		/*
3098 		 * First, add the delegation. (Although we must issue the
3099 		 * delegation, we can also ask for an immediate return.)
3100 		 */
3101 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3102 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
3103 		    clp->lc_clientid.lval[0];
3104 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
3105 		    clp->lc_clientid.lval[1];
3106 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
3107 		    nfsrv_nextstateindex(clp);
3108 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
3109 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3110 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3111 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3112 		    nfsrv_writedelegcnt++;
3113 		} else {
3114 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3115 			NFSLCK_READACCESS);
3116 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
3117 		}
3118 		new_deleg->ls_uid = new_stp->ls_uid;
3119 		new_deleg->ls_lfp = lfp;
3120 		new_deleg->ls_clp = clp;
3121 		new_deleg->ls_filerev = filerev;
3122 		new_deleg->ls_compref = nd->nd_compref;
3123 		new_deleg->ls_lastrecall = 0;
3124 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3125 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3126 		    new_deleg->ls_stateid), new_deleg, ls_hash);
3127 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3128 		new_deleg = NULL;
3129 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3130 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3131 		     LCL_CALLBACKSON ||
3132 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3133 		    !NFSVNO_DELEGOK(vp))
3134 		    *rflagsp |= NFSV4OPEN_RECALL;
3135 		NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
3136 		nfsrv_openpluslock++;
3137 		nfsrv_delegatecnt++;
3138 
3139 		/*
3140 		 * Now, do the associated open.
3141 		 */
3142 		new_open->ls_stateid.seqid = 1;
3143 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3144 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3145 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3146 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3147 		    NFSLCK_OPEN;
3148 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3149 			new_open->ls_flags |= (NFSLCK_READACCESS |
3150 			    NFSLCK_WRITEACCESS);
3151 		else
3152 			new_open->ls_flags |= NFSLCK_READACCESS;
3153 		new_open->ls_uid = new_stp->ls_uid;
3154 		new_open->ls_lfp = lfp;
3155 		new_open->ls_clp = clp;
3156 		LIST_INIT(&new_open->ls_open);
3157 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3158 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3159 		   new_open, ls_hash);
3160 		/*
3161 		 * and handle the open owner
3162 		 */
3163 		if (ownerstp) {
3164 		    new_open->ls_openowner = ownerstp;
3165 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3166 		} else {
3167 		    new_open->ls_openowner = new_stp;
3168 		    new_stp->ls_flags = 0;
3169 		    nfsrvd_refcache(new_stp->ls_op);
3170 		    new_stp->ls_noopens = 0;
3171 		    LIST_INIT(&new_stp->ls_open);
3172 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3173 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3174 		    *new_stpp = NULL;
3175 		    NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3176 		    nfsrv_openpluslock++;
3177 		}
3178 		openstp = new_open;
3179 		new_open = NULL;
3180 		NFSD_VNET(nfsstatsv1_p)->srvopens++;
3181 		nfsrv_openpluslock++;
3182 	    } else {
3183 		error = NFSERR_RECLAIMCONFLICT;
3184 	    }
3185 	} else if (ownerstp) {
3186 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3187 		    /* Replace the open */
3188 		    if (ownerstp->ls_op)
3189 			nfsrvd_derefcache(ownerstp->ls_op);
3190 		    ownerstp->ls_op = new_stp->ls_op;
3191 		    nfsrvd_refcache(ownerstp->ls_op);
3192 		    ownerstp->ls_seq = new_stp->ls_seq;
3193 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3194 		    stp = LIST_FIRST(&ownerstp->ls_open);
3195 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3196 			NFSLCK_OPEN;
3197 		    stp->ls_stateid.seqid = 1;
3198 		    stp->ls_uid = new_stp->ls_uid;
3199 		    if (lfp != stp->ls_lfp) {
3200 			LIST_REMOVE(stp, ls_file);
3201 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3202 			stp->ls_lfp = lfp;
3203 		    }
3204 		    openstp = stp;
3205 		} else if (openstp) {
3206 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3207 		    openstp->ls_stateid.seqid++;
3208 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3209 			openstp->ls_stateid.seqid == 0)
3210 			openstp->ls_stateid.seqid = 1;
3211 
3212 		    /*
3213 		     * This is where we can choose to issue a delegation.
3214 		     */
3215 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3216 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3217 		    else if (nfsrv_issuedelegs == 0)
3218 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3219 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3220 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3221 		    else if (delegate == 0 || writedeleg == 0 ||
3222 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3223 			nfsrv_writedelegifpos == 0) ||
3224 			!NFSVNO_DELEGOK(vp) ||
3225 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3226 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3227 			 LCL_CALLBACKSON)
3228 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3229 		    else {
3230 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3231 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3232 			    = clp->lc_clientid.lval[0];
3233 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3234 			    = clp->lc_clientid.lval[1];
3235 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3236 			    = nfsrv_nextstateindex(clp);
3237 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3238 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3239 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3240 			new_deleg->ls_uid = new_stp->ls_uid;
3241 			new_deleg->ls_lfp = lfp;
3242 			new_deleg->ls_clp = clp;
3243 			new_deleg->ls_filerev = filerev;
3244 			new_deleg->ls_compref = nd->nd_compref;
3245 			new_deleg->ls_lastrecall = 0;
3246 			nfsrv_writedelegcnt++;
3247 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3248 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3249 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3250 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3251 			new_deleg = NULL;
3252 			NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
3253 			nfsrv_openpluslock++;
3254 			nfsrv_delegatecnt++;
3255 		    }
3256 		} else {
3257 		    new_open->ls_stateid.seqid = 1;
3258 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3259 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3260 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3261 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3262 			NFSLCK_OPEN;
3263 		    new_open->ls_uid = new_stp->ls_uid;
3264 		    new_open->ls_openowner = ownerstp;
3265 		    new_open->ls_lfp = lfp;
3266 		    new_open->ls_clp = clp;
3267 		    LIST_INIT(&new_open->ls_open);
3268 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3269 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3270 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3271 			new_open, ls_hash);
3272 		    openstp = new_open;
3273 		    new_open = NULL;
3274 		    NFSD_VNET(nfsstatsv1_p)->srvopens++;
3275 		    nfsrv_openpluslock++;
3276 
3277 		    /*
3278 		     * This is where we can choose to issue a delegation.
3279 		     */
3280 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3281 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3282 		    else if (nfsrv_issuedelegs == 0)
3283 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3284 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3285 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3286 		    else if (delegate == 0 || (writedeleg == 0 &&
3287 			readonly == 0) || !NFSVNO_DELEGOK(vp) ||
3288 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3289 			 LCL_CALLBACKSON)
3290 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3291 		    else {
3292 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3293 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3294 			    = clp->lc_clientid.lval[0];
3295 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3296 			    = clp->lc_clientid.lval[1];
3297 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3298 			    = nfsrv_nextstateindex(clp);
3299 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3300 			    (nfsrv_writedelegifpos || !readonly) &&
3301 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3302 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3303 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3304 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3305 			    nfsrv_writedelegcnt++;
3306 			} else {
3307 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3308 				NFSLCK_READACCESS);
3309 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3310 			}
3311 			new_deleg->ls_uid = new_stp->ls_uid;
3312 			new_deleg->ls_lfp = lfp;
3313 			new_deleg->ls_clp = clp;
3314 			new_deleg->ls_filerev = filerev;
3315 			new_deleg->ls_compref = nd->nd_compref;
3316 			new_deleg->ls_lastrecall = 0;
3317 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3318 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3319 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3320 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3321 			new_deleg = NULL;
3322 			NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
3323 			nfsrv_openpluslock++;
3324 			nfsrv_delegatecnt++;
3325 		    }
3326 		}
3327 	} else {
3328 		/*
3329 		 * New owner case. Start the open_owner sequence with a
3330 		 * Needs confirmation (unless a reclaim) and hang the
3331 		 * new open off it.
3332 		 */
3333 		new_open->ls_stateid.seqid = 1;
3334 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3335 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3336 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3337 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3338 		    NFSLCK_OPEN;
3339 		new_open->ls_uid = new_stp->ls_uid;
3340 		LIST_INIT(&new_open->ls_open);
3341 		new_open->ls_openowner = new_stp;
3342 		new_open->ls_lfp = lfp;
3343 		new_open->ls_clp = clp;
3344 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3345 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3346 			new_stp->ls_flags = 0;
3347 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3348 			/* NFSv4.1 never needs confirmation. */
3349 			new_stp->ls_flags = 0;
3350 
3351 			/*
3352 			 * This is where we can choose to issue a delegation.
3353 			 */
3354 			if (delegate && nfsrv_issuedelegs &&
3355 			    (writedeleg || readonly) &&
3356 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3357 			     LCL_CALLBACKSON &&
3358 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3359 			    NFSVNO_DELEGOK(vp) &&
3360 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3361 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3362 				new_deleg->ls_stateid.seqid =
3363 				    delegstateidp->seqid = 1;
3364 				new_deleg->ls_stateid.other[0] =
3365 				    delegstateidp->other[0]
3366 				    = clp->lc_clientid.lval[0];
3367 				new_deleg->ls_stateid.other[1] =
3368 				    delegstateidp->other[1]
3369 				    = clp->lc_clientid.lval[1];
3370 				new_deleg->ls_stateid.other[2] =
3371 				    delegstateidp->other[2]
3372 				    = nfsrv_nextstateindex(clp);
3373 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3374 				    (nfsrv_writedelegifpos || !readonly) &&
3375 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3376 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3377 				     0)) {
3378 					new_deleg->ls_flags =
3379 					    (NFSLCK_DELEGWRITE |
3380 					     NFSLCK_READACCESS |
3381 					     NFSLCK_WRITEACCESS);
3382 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3383 					nfsrv_writedelegcnt++;
3384 				} else {
3385 					new_deleg->ls_flags =
3386 					    (NFSLCK_DELEGREAD |
3387 					     NFSLCK_READACCESS);
3388 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3389 				}
3390 				new_deleg->ls_uid = new_stp->ls_uid;
3391 				new_deleg->ls_lfp = lfp;
3392 				new_deleg->ls_clp = clp;
3393 				new_deleg->ls_filerev = filerev;
3394 				new_deleg->ls_compref = nd->nd_compref;
3395 				new_deleg->ls_lastrecall = 0;
3396 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3397 				    ls_file);
3398 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3399 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3400 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3401 				    ls_list);
3402 				new_deleg = NULL;
3403 				NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
3404 				nfsrv_openpluslock++;
3405 				nfsrv_delegatecnt++;
3406 			}
3407 			/*
3408 			 * Since NFSv4.1 never does an OpenConfirm, the first
3409 			 * open state will be acquired here.
3410 			 */
3411 			if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3412 				clp->lc_flags |= LCL_STAMPEDSTABLE;
3413 				len = clp->lc_idlen;
3414 				NFSBCOPY(clp->lc_id, clidp, len);
3415 				gotstate = 1;
3416 			}
3417 		} else {
3418 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3419 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3420 		}
3421 		nfsrvd_refcache(new_stp->ls_op);
3422 		new_stp->ls_noopens = 0;
3423 		LIST_INIT(&new_stp->ls_open);
3424 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3425 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3426 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3427 		    new_open, ls_hash);
3428 		openstp = new_open;
3429 		new_open = NULL;
3430 		*new_stpp = NULL;
3431 		NFSD_VNET(nfsstatsv1_p)->srvopens++;
3432 		nfsrv_openpluslock++;
3433 		NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3434 		nfsrv_openpluslock++;
3435 	}
3436 	if (!error) {
3437 		stateidp->seqid = openstp->ls_stateid.seqid;
3438 		stateidp->other[0] = openstp->ls_stateid.other[0];
3439 		stateidp->other[1] = openstp->ls_stateid.other[1];
3440 		stateidp->other[2] = openstp->ls_stateid.other[2];
3441 	}
3442 	NFSUNLOCKSTATE();
3443 	if (haslock) {
3444 		NFSLOCKV4ROOTMUTEX();
3445 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3446 		NFSUNLOCKV4ROOTMUTEX();
3447 	}
3448 	if (new_open)
3449 		free(new_open, M_NFSDSTATE);
3450 	if (new_deleg)
3451 		free(new_deleg, M_NFSDSTATE);
3452 
3453 	/*
3454 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3455 	 * to the stable storage file.
3456 	 */
3457 	if (gotstate != 0) {
3458 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3459 		nfsrv_backupstable();
3460 	}
3461 
3462 out:
3463 	free(clidp, M_TEMP);
3464 	NFSEXITCODE2(error, nd);
3465 	return (error);
3466 }
3467 
3468 /*
3469  * Open update. Does the confirm, downgrade and close.
3470  */
3471 int
3472 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3473     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p,
3474     int *retwriteaccessp)
3475 {
3476 	struct nfsstate *stp;
3477 	struct nfsclient *clp;
3478 	struct nfslockfile *lfp;
3479 	u_int32_t bits;
3480 	int error = 0, gotstate = 0, len = 0;
3481 	u_char *clidp = NULL;
3482 
3483 	/*
3484 	 * Check for restart conditions (client and server).
3485 	 */
3486 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3487 	    &new_stp->ls_stateid, 0);
3488 	if (error)
3489 		goto out;
3490 
3491 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3492 	NFSLOCKSTATE();
3493 	/*
3494 	 * Get the open structure via clientid and stateid.
3495 	 */
3496 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3497 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3498 	if (!error)
3499 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3500 		    new_stp->ls_flags, &stp);
3501 
3502 	/*
3503 	 * Sanity check the open.
3504 	 */
3505 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3506 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3507 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3508 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3509 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3510 		error = NFSERR_BADSTATEID;
3511 
3512 	if (!error)
3513 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3514 		    stp->ls_openowner, new_stp->ls_op);
3515 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3516 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3517 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3518 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3519 	      new_stp->ls_stateid.seqid != 0)))
3520 		error = NFSERR_OLDSTATEID;
3521 	if (!error && vp->v_type != VREG) {
3522 		if (vp->v_type == VDIR)
3523 			error = NFSERR_ISDIR;
3524 		else
3525 			error = NFSERR_INVAL;
3526 	}
3527 
3528 	if (error) {
3529 		/*
3530 		 * If a client tries to confirm an Open with a bad
3531 		 * seqid# and there are no byte range locks or other Opens
3532 		 * on the openowner, just throw it away, so the next use of the
3533 		 * openowner will start a fresh seq#.
3534 		 */
3535 		if (error == NFSERR_BADSEQID &&
3536 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3537 		    nfsrv_nootherstate(stp))
3538 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3539 		NFSUNLOCKSTATE();
3540 		goto out;
3541 	}
3542 
3543 	/*
3544 	 * Set the return stateid.
3545 	 */
3546 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3547 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3548 		stateidp->seqid = 1;
3549 	stateidp->other[0] = stp->ls_stateid.other[0];
3550 	stateidp->other[1] = stp->ls_stateid.other[1];
3551 	stateidp->other[2] = stp->ls_stateid.other[2];
3552 	/*
3553 	 * Now, handle the three cases.
3554 	 */
3555 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3556 		/*
3557 		 * If the open doesn't need confirmation, it seems to me that
3558 		 * there is a client error, but I'll just log it and keep going?
3559 		 */
3560 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3561 			printf("Nfsv4d: stray open confirm\n");
3562 		stp->ls_openowner->ls_flags = 0;
3563 		stp->ls_stateid.seqid++;
3564 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3565 		    stp->ls_stateid.seqid == 0)
3566 			stp->ls_stateid.seqid = 1;
3567 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3568 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3569 			len = clp->lc_idlen;
3570 			NFSBCOPY(clp->lc_id, clidp, len);
3571 			gotstate = 1;
3572 		}
3573 		NFSUNLOCKSTATE();
3574 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3575 		lfp = stp->ls_lfp;
3576 		if (retwriteaccessp != NULL) {
3577 			if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0)
3578 				*retwriteaccessp = 1;
3579 			else
3580 				*retwriteaccessp = 0;
3581 		}
3582 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3583 			/* Get the lf lock */
3584 			nfsrv_locklf(lfp);
3585 			NFSUNLOCKSTATE();
3586 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3587 			NFSVOPUNLOCK(vp);
3588 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3589 				NFSLOCKSTATE();
3590 				nfsrv_unlocklf(lfp);
3591 				NFSUNLOCKSTATE();
3592 			}
3593 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3594 		} else {
3595 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3596 			NFSUNLOCKSTATE();
3597 		}
3598 	} else {
3599 		/*
3600 		 * Update the share bits, making sure that the new set are a
3601 		 * subset of the old ones.
3602 		 */
3603 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3604 		if (~(stp->ls_flags) & bits) {
3605 			NFSUNLOCKSTATE();
3606 			error = NFSERR_INVAL;
3607 			goto out;
3608 		}
3609 		stp->ls_flags = (bits | NFSLCK_OPEN);
3610 		stp->ls_stateid.seqid++;
3611 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3612 		    stp->ls_stateid.seqid == 0)
3613 			stp->ls_stateid.seqid = 1;
3614 		NFSUNLOCKSTATE();
3615 	}
3616 
3617 	/*
3618 	 * If the client just confirmed its first open, write a timestamp
3619 	 * to the stable storage file.
3620 	 */
3621 	if (gotstate != 0) {
3622 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3623 		nfsrv_backupstable();
3624 	}
3625 
3626 out:
3627 	free(clidp, M_TEMP);
3628 	NFSEXITCODE2(error, nd);
3629 	return (error);
3630 }
3631 
3632 /*
3633  * Delegation update. Does the purge and return.
3634  */
3635 int
3636 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3637     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3638     NFSPROC_T *p, int *retwriteaccessp)
3639 {
3640 	struct nfsstate *stp;
3641 	struct nfsclient *clp;
3642 	int error = 0;
3643 	fhandle_t fh;
3644 
3645 	/*
3646 	 * Do a sanity check against the file handle for DelegReturn.
3647 	 */
3648 	if (vp) {
3649 		error = nfsvno_getfh(vp, &fh, p);
3650 		if (error)
3651 			goto out;
3652 	}
3653 	/*
3654 	 * Check for restart conditions (client and server).
3655 	 */
3656 	if (op == NFSV4OP_DELEGRETURN)
3657 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3658 			stateidp, 0);
3659 	else
3660 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3661 			stateidp, 0);
3662 
3663 	NFSLOCKSTATE();
3664 	/*
3665 	 * Get the open structure via clientid and stateid.
3666 	 */
3667 	if (!error)
3668 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3669 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3670 	if (error) {
3671 		if (error == NFSERR_CBPATHDOWN)
3672 			error = 0;
3673 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3674 			error = NFSERR_STALESTATEID;
3675 	}
3676 	if (!error && op == NFSV4OP_DELEGRETURN) {
3677 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3678 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3679 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3680 		error = NFSERR_OLDSTATEID;
3681 	}
3682 	/*
3683 	 * NFSERR_EXPIRED means that the state has gone away,
3684 	 * so Delegations have been purged. Just return ok.
3685 	 */
3686 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3687 		NFSUNLOCKSTATE();
3688 		error = 0;
3689 		goto out;
3690 	}
3691 	if (error) {
3692 		NFSUNLOCKSTATE();
3693 		goto out;
3694 	}
3695 
3696 	if (op == NFSV4OP_DELEGRETURN) {
3697 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3698 		    sizeof (fhandle_t))) {
3699 			NFSUNLOCKSTATE();
3700 			error = NFSERR_BADSTATEID;
3701 			goto out;
3702 		}
3703 		if (retwriteaccessp != NULL) {
3704 			if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
3705 				*retwriteaccessp = 1;
3706 			else
3707 				*retwriteaccessp = 0;
3708 		}
3709 		nfsrv_freedeleg(stp);
3710 	} else {
3711 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3712 	}
3713 	NFSUNLOCKSTATE();
3714 	error = 0;
3715 
3716 out:
3717 	NFSEXITCODE(error);
3718 	return (error);
3719 }
3720 
3721 /*
3722  * Release lock owner.
3723  */
3724 int
3725 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3726     NFSPROC_T *p)
3727 {
3728 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3729 	struct nfsclient *clp;
3730 	int error = 0;
3731 
3732 	/*
3733 	 * Check for restart conditions (client and server).
3734 	 */
3735 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3736 	    &new_stp->ls_stateid, 0);
3737 	if (error)
3738 		goto out;
3739 
3740 	NFSLOCKSTATE();
3741 	/*
3742 	 * Get the lock owner by name.
3743 	 */
3744 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3745 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3746 	if (error) {
3747 		NFSUNLOCKSTATE();
3748 		goto out;
3749 	}
3750 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3751 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3752 		stp = LIST_FIRST(&openstp->ls_open);
3753 		while (stp != LIST_END(&openstp->ls_open)) {
3754 		    nstp = LIST_NEXT(stp, ls_list);
3755 		    /*
3756 		     * If the owner matches, check for locks and
3757 		     * then free or return an error.
3758 		     */
3759 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3760 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3761 			 stp->ls_ownerlen)){
3762 			if (LIST_EMPTY(&stp->ls_lock)) {
3763 			    nfsrv_freelockowner(stp, NULL, 0, p);
3764 			} else {
3765 			    NFSUNLOCKSTATE();
3766 			    error = NFSERR_LOCKSHELD;
3767 			    goto out;
3768 			}
3769 		    }
3770 		    stp = nstp;
3771 		}
3772 	    }
3773 	}
3774 	NFSUNLOCKSTATE();
3775 
3776 out:
3777 	NFSEXITCODE(error);
3778 	return (error);
3779 }
3780 
3781 /*
3782  * Get the file handle for a lock structure.
3783  */
3784 static int
3785 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3786     fhandle_t *nfhp, NFSPROC_T *p)
3787 {
3788 	fhandle_t *fhp = NULL;
3789 	int error;
3790 
3791 	/*
3792 	 * For lock, use the new nfslock structure, otherwise just
3793 	 * a fhandle_t on the stack.
3794 	 */
3795 	if (flags & NFSLCK_OPEN) {
3796 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3797 		fhp = &new_lfp->lf_fh;
3798 	} else if (nfhp) {
3799 		fhp = nfhp;
3800 	} else {
3801 		panic("nfsrv_getlockfh");
3802 	}
3803 	error = nfsvno_getfh(vp, fhp, p);
3804 	NFSEXITCODE(error);
3805 	return (error);
3806 }
3807 
3808 /*
3809  * Get an nfs lock structure. Allocate one, as required, and return a
3810  * pointer to it.
3811  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3812  */
3813 static int
3814 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3815     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3816 {
3817 	struct nfslockfile *lfp;
3818 	fhandle_t *fhp = NULL, *tfhp;
3819 	struct nfslockhashhead *hp;
3820 	struct nfslockfile *new_lfp = NULL;
3821 
3822 	/*
3823 	 * For lock, use the new nfslock structure, otherwise just
3824 	 * a fhandle_t on the stack.
3825 	 */
3826 	if (flags & NFSLCK_OPEN) {
3827 		new_lfp = *new_lfpp;
3828 		fhp = &new_lfp->lf_fh;
3829 	} else if (nfhp) {
3830 		fhp = nfhp;
3831 	} else {
3832 		panic("nfsrv_getlockfile");
3833 	}
3834 
3835 	hp = NFSLOCKHASH(fhp);
3836 	LIST_FOREACH(lfp, hp, lf_hash) {
3837 		tfhp = &lfp->lf_fh;
3838 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3839 			if (lockit)
3840 				nfsrv_locklf(lfp);
3841 			*lfpp = lfp;
3842 			return (0);
3843 		}
3844 	}
3845 	if (!(flags & NFSLCK_OPEN))
3846 		return (-1);
3847 
3848 	/*
3849 	 * No match, so chain the new one into the list.
3850 	 */
3851 	LIST_INIT(&new_lfp->lf_open);
3852 	LIST_INIT(&new_lfp->lf_lock);
3853 	LIST_INIT(&new_lfp->lf_deleg);
3854 	LIST_INIT(&new_lfp->lf_locallock);
3855 	LIST_INIT(&new_lfp->lf_rollback);
3856 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3857 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3858 	new_lfp->lf_usecount = 0;
3859 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3860 	*lfpp = new_lfp;
3861 	*new_lfpp = NULL;
3862 	return (0);
3863 }
3864 
3865 /*
3866  * This function adds a nfslock lock structure to the list for the associated
3867  * nfsstate and nfslockfile structures. It will be inserted after the
3868  * entry pointed at by insert_lop.
3869  */
3870 static void
3871 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3872     struct nfsstate *stp, struct nfslockfile *lfp)
3873 {
3874 	struct nfslock *lop, *nlop;
3875 
3876 	new_lop->lo_stp = stp;
3877 	new_lop->lo_lfp = lfp;
3878 
3879 	if (stp != NULL) {
3880 		/* Insert in increasing lo_first order */
3881 		lop = LIST_FIRST(&lfp->lf_lock);
3882 		if (lop == LIST_END(&lfp->lf_lock) ||
3883 		    new_lop->lo_first <= lop->lo_first) {
3884 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3885 		} else {
3886 			nlop = LIST_NEXT(lop, lo_lckfile);
3887 			while (nlop != LIST_END(&lfp->lf_lock) &&
3888 			       nlop->lo_first < new_lop->lo_first) {
3889 				lop = nlop;
3890 				nlop = LIST_NEXT(lop, lo_lckfile);
3891 			}
3892 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3893 		}
3894 	} else {
3895 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3896 	}
3897 
3898 	/*
3899 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3900 	 * an empty list.
3901 	 */
3902 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3903 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3904 	else if ((struct nfsstate *)insert_lop == stp)
3905 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3906 	else
3907 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3908 	if (stp != NULL) {
3909 		NFSD_VNET(nfsstatsv1_p)->srvlocks++;
3910 		nfsrv_openpluslock++;
3911 	}
3912 }
3913 
3914 /*
3915  * This function updates the locking for a lock owner and given file. It
3916  * maintains a list of lock ranges ordered on increasing file offset that
3917  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3918  * It always adds new_lop to the list and sometimes uses the one pointed
3919  * at by other_lopp.
3920  */
3921 static void
3922 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3923     struct nfslock **other_lopp, struct nfslockfile *lfp)
3924 {
3925 	struct nfslock *new_lop = *new_lopp;
3926 	struct nfslock *lop, *tlop, *ilop;
3927 	struct nfslock *other_lop = *other_lopp;
3928 	int unlock = 0, myfile = 0;
3929 	u_int64_t tmp;
3930 
3931 	/*
3932 	 * Work down the list until the lock is merged.
3933 	 */
3934 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3935 		unlock = 1;
3936 	if (stp != NULL) {
3937 		ilop = (struct nfslock *)stp;
3938 		lop = LIST_FIRST(&stp->ls_lock);
3939 	} else {
3940 		ilop = (struct nfslock *)lfp;
3941 		lop = LIST_FIRST(&lfp->lf_locallock);
3942 	}
3943 	while (lop != NULL) {
3944 	    /*
3945 	     * Only check locks for this file that aren't before the start of
3946 	     * new lock's range.
3947 	     */
3948 	    if (lop->lo_lfp == lfp) {
3949 	      myfile = 1;
3950 	      if (lop->lo_end >= new_lop->lo_first) {
3951 		if (new_lop->lo_end < lop->lo_first) {
3952 			/*
3953 			 * If the new lock ends before the start of the
3954 			 * current lock's range, no merge, just insert
3955 			 * the new lock.
3956 			 */
3957 			break;
3958 		}
3959 		if (new_lop->lo_flags == lop->lo_flags ||
3960 		    (new_lop->lo_first <= lop->lo_first &&
3961 		     new_lop->lo_end >= lop->lo_end)) {
3962 			/*
3963 			 * This lock can be absorbed by the new lock/unlock.
3964 			 * This happens when it covers the entire range
3965 			 * of the old lock or is contiguous
3966 			 * with the old lock and is of the same type or an
3967 			 * unlock.
3968 			 */
3969 			if (lop->lo_first < new_lop->lo_first)
3970 				new_lop->lo_first = lop->lo_first;
3971 			if (lop->lo_end > new_lop->lo_end)
3972 				new_lop->lo_end = lop->lo_end;
3973 			tlop = lop;
3974 			lop = LIST_NEXT(lop, lo_lckowner);
3975 			nfsrv_freenfslock(tlop);
3976 			continue;
3977 		}
3978 
3979 		/*
3980 		 * All these cases are for contiguous locks that are not the
3981 		 * same type, so they can't be merged.
3982 		 */
3983 		if (new_lop->lo_first <= lop->lo_first) {
3984 			/*
3985 			 * This case is where the new lock overlaps with the
3986 			 * first part of the old lock. Move the start of the
3987 			 * old lock to just past the end of the new lock. The
3988 			 * new lock will be inserted in front of the old, since
3989 			 * ilop hasn't been updated. (We are done now.)
3990 			 */
3991 			lop->lo_first = new_lop->lo_end;
3992 			break;
3993 		}
3994 		if (new_lop->lo_end >= lop->lo_end) {
3995 			/*
3996 			 * This case is where the new lock overlaps with the
3997 			 * end of the old lock's range. Move the old lock's
3998 			 * end to just before the new lock's first and insert
3999 			 * the new lock after the old lock.
4000 			 * Might not be done yet, since the new lock could
4001 			 * overlap further locks with higher ranges.
4002 			 */
4003 			lop->lo_end = new_lop->lo_first;
4004 			ilop = lop;
4005 			lop = LIST_NEXT(lop, lo_lckowner);
4006 			continue;
4007 		}
4008 		/*
4009 		 * The final case is where the new lock's range is in the
4010 		 * middle of the current lock's and splits the current lock
4011 		 * up. Use *other_lopp to handle the second part of the
4012 		 * split old lock range. (We are done now.)
4013 		 * For unlock, we use new_lop as other_lop and tmp, since
4014 		 * other_lop and new_lop are the same for this case.
4015 		 * We noted the unlock case above, so we don't need
4016 		 * new_lop->lo_flags any longer.
4017 		 */
4018 		tmp = new_lop->lo_first;
4019 		if (other_lop == NULL) {
4020 			if (!unlock)
4021 				panic("nfsd srv update unlock");
4022 			other_lop = new_lop;
4023 			*new_lopp = NULL;
4024 		}
4025 		other_lop->lo_first = new_lop->lo_end;
4026 		other_lop->lo_end = lop->lo_end;
4027 		other_lop->lo_flags = lop->lo_flags;
4028 		other_lop->lo_stp = stp;
4029 		other_lop->lo_lfp = lfp;
4030 		lop->lo_end = tmp;
4031 		nfsrv_insertlock(other_lop, lop, stp, lfp);
4032 		*other_lopp = NULL;
4033 		ilop = lop;
4034 		break;
4035 	      }
4036 	    }
4037 	    ilop = lop;
4038 	    lop = LIST_NEXT(lop, lo_lckowner);
4039 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
4040 		break;
4041 	}
4042 
4043 	/*
4044 	 * Insert the new lock in the list at the appropriate place.
4045 	 */
4046 	if (!unlock) {
4047 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
4048 		*new_lopp = NULL;
4049 	}
4050 }
4051 
4052 /*
4053  * This function handles sequencing of locks, etc.
4054  * It returns an error that indicates what the caller should do.
4055  */
4056 static int
4057 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
4058     struct nfsstate *stp, struct nfsrvcache *op)
4059 {
4060 	int error = 0;
4061 
4062 	if ((nd->nd_flag & ND_NFSV41) != 0)
4063 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
4064 		goto out;
4065 	if (op != nd->nd_rp)
4066 		panic("nfsrvstate checkseqid");
4067 	if (!(op->rc_flag & RC_INPROG))
4068 		panic("nfsrvstate not inprog");
4069 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
4070 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
4071 		panic("nfsrvstate op refcnt");
4072 	}
4073 
4074 	/* If ND_ERELOOKUP is set, the seqid has already been handled. */
4075 	if ((nd->nd_flag & ND_ERELOOKUP) != 0)
4076 		goto out;
4077 
4078 	if ((stp->ls_seq + 1) == seqid) {
4079 		if (stp->ls_op)
4080 			nfsrvd_derefcache(stp->ls_op);
4081 		stp->ls_op = op;
4082 		nfsrvd_refcache(op);
4083 		stp->ls_seq = seqid;
4084 		goto out;
4085 	} else if (stp->ls_seq == seqid && stp->ls_op &&
4086 		op->rc_xid == stp->ls_op->rc_xid &&
4087 		op->rc_refcnt == 0 &&
4088 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
4089 		op->rc_cksum == stp->ls_op->rc_cksum) {
4090 		if (stp->ls_op->rc_flag & RC_INPROG) {
4091 			error = NFSERR_DONTREPLY;
4092 			goto out;
4093 		}
4094 		nd->nd_rp = stp->ls_op;
4095 		nd->nd_rp->rc_flag |= RC_INPROG;
4096 		nfsrvd_delcache(op);
4097 		error = NFSERR_REPLYFROMCACHE;
4098 		goto out;
4099 	}
4100 	error = NFSERR_BADSEQID;
4101 
4102 out:
4103 	NFSEXITCODE2(error, nd);
4104 	return (error);
4105 }
4106 
4107 /*
4108  * Get the client ip address for callbacks. If the strings can't be parsed,
4109  * just set lc_program to 0 to indicate no callbacks are possible.
4110  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
4111  *  the address to the client's transport address. This won't be used
4112  *  for callbacks, but can be printed out by nfsstats for info.)
4113  * Return error if the xdr can't be parsed, 0 otherwise.
4114  */
4115 int
4116 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
4117 {
4118 	u_int32_t *tl;
4119 	u_char *cp, *cp2;
4120 	int i, j, maxalen = 0, minalen = 0;
4121 	sa_family_t af;
4122 #ifdef INET
4123 	struct sockaddr_in *rin = NULL, *sin;
4124 #endif
4125 #ifdef INET6
4126 	struct sockaddr_in6 *rin6 = NULL, *sin6;
4127 #endif
4128 	u_char *addr;
4129 	int error = 0, cantparse = 0;
4130 	union {
4131 		in_addr_t ival;
4132 		u_char cval[4];
4133 	} ip;
4134 	union {
4135 		in_port_t sval;
4136 		u_char cval[2];
4137 	} port;
4138 
4139 	/* 8 is the maximum length of the port# string. */
4140 	addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK);
4141 	clp->lc_req.nr_client = NULL;
4142 	clp->lc_req.nr_lock = 0;
4143 	af = AF_UNSPEC;
4144 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4145 	i = fxdr_unsigned(int, *tl);
4146 	if (i >= 3 && i <= 4) {
4147 		error = nfsrv_mtostr(nd, addr, i);
4148 		if (error)
4149 			goto nfsmout;
4150 #ifdef INET
4151 		if (!strcmp(addr, "tcp")) {
4152 			clp->lc_flags |= LCL_TCPCALLBACK;
4153 			clp->lc_req.nr_sotype = SOCK_STREAM;
4154 			clp->lc_req.nr_soproto = IPPROTO_TCP;
4155 			af = AF_INET;
4156 		} else if (!strcmp(addr, "udp")) {
4157 			clp->lc_req.nr_sotype = SOCK_DGRAM;
4158 			clp->lc_req.nr_soproto = IPPROTO_UDP;
4159 			af = AF_INET;
4160 		}
4161 #endif
4162 #ifdef INET6
4163 		if (af == AF_UNSPEC) {
4164 			if (!strcmp(addr, "tcp6")) {
4165 				clp->lc_flags |= LCL_TCPCALLBACK;
4166 				clp->lc_req.nr_sotype = SOCK_STREAM;
4167 				clp->lc_req.nr_soproto = IPPROTO_TCP;
4168 				af = AF_INET6;
4169 			} else if (!strcmp(addr, "udp6")) {
4170 				clp->lc_req.nr_sotype = SOCK_DGRAM;
4171 				clp->lc_req.nr_soproto = IPPROTO_UDP;
4172 				af = AF_INET6;
4173 			}
4174 		}
4175 #endif
4176 		if (af == AF_UNSPEC) {
4177 			cantparse = 1;
4178 		}
4179 	} else {
4180 		cantparse = 1;
4181 		if (i > 0) {
4182 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4183 			if (error)
4184 				goto nfsmout;
4185 		}
4186 	}
4187 	/*
4188 	 * The caller has allocated clp->lc_req.nr_nam to be large enough
4189 	 * for either AF_INET or AF_INET6 and zeroed out the contents.
4190 	 * maxalen is set to the maximum length of the host IP address string
4191 	 * plus 8 for the maximum length of the port#.
4192 	 * minalen is set to the minimum length of the host IP address string
4193 	 * plus 4 for the minimum length of the port#.
4194 	 * These lengths do not include NULL termination,
4195 	 * so INET[6]_ADDRSTRLEN - 1 is used in the calculations.
4196 	 */
4197 	switch (af) {
4198 #ifdef INET
4199 	case AF_INET:
4200 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4201 		rin->sin_family = AF_INET;
4202 		rin->sin_len = sizeof(struct sockaddr_in);
4203 		maxalen = INET_ADDRSTRLEN - 1 + 8;
4204 		minalen = 7 + 4;
4205 		break;
4206 #endif
4207 #ifdef INET6
4208 	case AF_INET6:
4209 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4210 		rin6->sin6_family = AF_INET6;
4211 		rin6->sin6_len = sizeof(struct sockaddr_in6);
4212 		maxalen = INET6_ADDRSTRLEN - 1 + 8;
4213 		minalen = 3 + 4;
4214 		break;
4215 #endif
4216 	}
4217 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4218 	i = fxdr_unsigned(int, *tl);
4219 	if (i < 0) {
4220 		error = NFSERR_BADXDR;
4221 		goto nfsmout;
4222 	} else if (i == 0) {
4223 		cantparse = 1;
4224 	} else if (!cantparse && i <= maxalen && i >= minalen) {
4225 		error = nfsrv_mtostr(nd, addr, i);
4226 		if (error)
4227 			goto nfsmout;
4228 
4229 		/*
4230 		 * Parse out the address fields. We expect 6 decimal numbers
4231 		 * separated by '.'s for AF_INET and two decimal numbers
4232 		 * preceeded by '.'s for AF_INET6.
4233 		 */
4234 		cp = NULL;
4235 		switch (af) {
4236 #ifdef INET6
4237 		/*
4238 		 * For AF_INET6, first parse the host address.
4239 		 */
4240 		case AF_INET6:
4241 			cp = strchr(addr, '.');
4242 			if (cp != NULL) {
4243 				*cp++ = '\0';
4244 				if (inet_pton(af, addr, &rin6->sin6_addr) == 1)
4245 					i = 4;
4246 				else {
4247 					cp = NULL;
4248 					cantparse = 1;
4249 				}
4250 			}
4251 			break;
4252 #endif
4253 #ifdef INET
4254 		case AF_INET:
4255 			cp = addr;
4256 			i = 0;
4257 			break;
4258 #endif
4259 		}
4260 		while (cp != NULL && *cp && i < 6) {
4261 			cp2 = cp;
4262 			while (*cp2 && *cp2 != '.')
4263 				cp2++;
4264 			if (*cp2)
4265 				*cp2++ = '\0';
4266 			else if (i != 5) {
4267 				cantparse = 1;
4268 				break;
4269 			}
4270 			j = nfsrv_getipnumber(cp);
4271 			if (j >= 0) {
4272 				if (i < 4)
4273 					ip.cval[3 - i] = j;
4274 				else
4275 					port.cval[5 - i] = j;
4276 			} else {
4277 				cantparse = 1;
4278 				break;
4279 			}
4280 			cp = cp2;
4281 			i++;
4282 		}
4283 		if (!cantparse) {
4284 			/*
4285 			 * The host address INADDR_ANY is (mis)used to indicate
4286 			 * "there is no valid callback address".
4287 			 */
4288 			switch (af) {
4289 #ifdef INET6
4290 			case AF_INET6:
4291 				if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr,
4292 				    &in6addr_any))
4293 					rin6->sin6_port = htons(port.sval);
4294 				else
4295 					cantparse = 1;
4296 				break;
4297 #endif
4298 #ifdef INET
4299 			case AF_INET:
4300 				if (ip.ival != INADDR_ANY) {
4301 					rin->sin_addr.s_addr = htonl(ip.ival);
4302 					rin->sin_port = htons(port.sval);
4303 				} else {
4304 					cantparse = 1;
4305 				}
4306 				break;
4307 #endif
4308 			}
4309 		}
4310 	} else {
4311 		cantparse = 1;
4312 		if (i > 0) {
4313 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4314 			if (error)
4315 				goto nfsmout;
4316 		}
4317 	}
4318 	if (cantparse) {
4319 		switch (nd->nd_nam->sa_family) {
4320 #ifdef INET
4321 		case AF_INET:
4322 			sin = (struct sockaddr_in *)nd->nd_nam;
4323 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4324 			rin->sin_family = AF_INET;
4325 			rin->sin_len = sizeof(struct sockaddr_in);
4326 			rin->sin_addr.s_addr = sin->sin_addr.s_addr;
4327 			rin->sin_port = 0x0;
4328 			break;
4329 #endif
4330 #ifdef INET6
4331 		case AF_INET6:
4332 			sin6 = (struct sockaddr_in6 *)nd->nd_nam;
4333 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4334 			rin6->sin6_family = AF_INET6;
4335 			rin6->sin6_len = sizeof(struct sockaddr_in6);
4336 			rin6->sin6_addr = sin6->sin6_addr;
4337 			rin6->sin6_port = 0x0;
4338 			break;
4339 #endif
4340 		}
4341 		clp->lc_program = 0;
4342 	}
4343 nfsmout:
4344 	free(addr, M_TEMP);
4345 	NFSEXITCODE2(error, nd);
4346 	return (error);
4347 }
4348 
4349 /*
4350  * Turn a string of up to three decimal digits into a number. Return -1 upon
4351  * error.
4352  */
4353 static int
4354 nfsrv_getipnumber(u_char *cp)
4355 {
4356 	int i = 0, j = 0;
4357 
4358 	while (*cp) {
4359 		if (j > 2 || *cp < '0' || *cp > '9')
4360 			return (-1);
4361 		i *= 10;
4362 		i += (*cp - '0');
4363 		cp++;
4364 		j++;
4365 	}
4366 	if (i < 256)
4367 		return (i);
4368 	return (-1);
4369 }
4370 
4371 /*
4372  * This function checks for restart conditions.
4373  */
4374 static int
4375 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4376     nfsv4stateid_t *stateidp, int specialid)
4377 {
4378 	int ret = 0;
4379 
4380 	/*
4381 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4382 	 * and DelegPurge have a clientid, the rest a stateid.
4383 	 */
4384 	if (flags &
4385 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4386 		if (clientid.lval[0] != NFSD_VNET(nfsrvboottime)) {
4387 			ret = NFSERR_STALECLIENTID;
4388 			goto out;
4389 		}
4390 	} else if (stateidp->other[0] != NFSD_VNET(nfsrvboottime) &&
4391 		specialid == 0) {
4392 		ret = NFSERR_STALESTATEID;
4393 		goto out;
4394 	}
4395 
4396 	/*
4397 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4398 	 * not use a lock/open owner seqid#, so the check can be done now.
4399 	 * (The others will be checked, as required, later.)
4400 	 */
4401 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4402 		goto out;
4403 
4404 	NFSLOCKSTATE();
4405 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4406 	NFSUNLOCKSTATE();
4407 
4408 out:
4409 	NFSEXITCODE(ret);
4410 	return (ret);
4411 }
4412 
4413 /*
4414  * Check for grace.
4415  */
4416 static int
4417 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4418     u_int32_t flags)
4419 {
4420 	int error = 0, notreclaimed;
4421 	struct nfsrv_stable *sp;
4422 
4423 	if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & (NFSNSF_UPDATEDONE |
4424 	     NFSNSF_GRACEOVER)) == 0) {
4425 		/*
4426 		 * First, check to see if all of the clients have done a
4427 		 * ReclaimComplete.  If so, grace can end now.
4428 		 */
4429 		notreclaimed = 0;
4430 		LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head,
4431 		    nst_list) {
4432 			if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4433 				notreclaimed = 1;
4434 				break;
4435 			}
4436 		}
4437 		if (notreclaimed == 0)
4438 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
4439 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4440 	}
4441 
4442 	if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_GRACEOVER) != 0) {
4443 		if (flags & NFSLCK_RECLAIM) {
4444 			error = NFSERR_NOGRACE;
4445 			goto out;
4446 		}
4447 	} else {
4448 		if (!(flags & NFSLCK_RECLAIM)) {
4449 			error = NFSERR_GRACE;
4450 			goto out;
4451 		}
4452 		if (nd != NULL && clp != NULL &&
4453 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4454 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4455 			error = NFSERR_NOGRACE;
4456 			goto out;
4457 		}
4458 
4459 		/*
4460 		 * If grace is almost over and we are still getting Reclaims,
4461 		 * extend grace a bit.
4462 		 */
4463 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4464 		    NFSD_VNET(nfsrv_stablefirst).nsf_eograce)
4465 			NFSD_VNET(nfsrv_stablefirst).nsf_eograce =
4466 				NFSD_MONOSEC + NFSRV_LEASEDELTA;
4467 	}
4468 
4469 out:
4470 	NFSEXITCODE(error);
4471 	return (error);
4472 }
4473 
4474 /*
4475  * Do a server callback.
4476  * The "trunc" argument is slightly overloaded and refers to different
4477  * boolean arguments for CBRECALL and CBLAYOUTRECALL.
4478  */
4479 static int
4480 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp,
4481     int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp,
4482     int laytype, NFSPROC_T *p)
4483 {
4484 	struct mbuf *m;
4485 	u_int32_t *tl;
4486 	struct nfsrv_descript *nd;
4487 	struct ucred *cred;
4488 	int error = 0, slotpos;
4489 	u_int32_t callback;
4490 	struct nfsdsession *sep = NULL;
4491 	uint64_t tval;
4492 	bool dotls;
4493 
4494 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4495 	cred = newnfs_getcred();
4496 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4497 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4498 		NFSUNLOCKSTATE();
4499 		panic("docallb");
4500 	}
4501 	clp->lc_cbref++;
4502 
4503 	/*
4504 	 * Fill the callback program# and version into the request
4505 	 * structure for newnfs_connect() to use.
4506 	 */
4507 	clp->lc_req.nr_prog = clp->lc_program;
4508 #ifdef notnow
4509 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4510 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4511 	else
4512 #endif
4513 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4514 
4515 	/*
4516 	 * First, fill in some of the fields of nd and cr.
4517 	 */
4518 	nd->nd_flag = ND_NFSV4;
4519 	if (clp->lc_flags & LCL_GSS)
4520 		nd->nd_flag |= ND_KERBV;
4521 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4522 		nd->nd_flag |= ND_NFSV41;
4523 	if ((clp->lc_flags & LCL_NFSV42) != 0)
4524 		nd->nd_flag |= ND_NFSV42;
4525 	nd->nd_repstat = 0;
4526 	cred->cr_uid = clp->lc_uid;
4527 	cred->cr_gid = clp->lc_gid;
4528 	callback = clp->lc_callback;
4529 	NFSUNLOCKSTATE();
4530 	cred->cr_ngroups = 1;
4531 
4532 	/*
4533 	 * Get the first mbuf for the request.
4534 	 */
4535 	MGET(m, M_WAITOK, MT_DATA);
4536 	m->m_len = 0;
4537 	nd->nd_mreq = nd->nd_mb = m;
4538 	nd->nd_bpos = mtod(m, caddr_t);
4539 
4540 	/*
4541 	 * and build the callback request.
4542 	 */
4543 	if (procnum == NFSV4OP_CBGETATTR) {
4544 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4545 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4546 		    "CB Getattr", &sep, &slotpos);
4547 		if (error != 0) {
4548 			m_freem(nd->nd_mreq);
4549 			goto errout;
4550 		}
4551 		(void)nfsm_fhtom(NULL, nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4552 		(void)nfsrv_putattrbit(nd, attrbitp);
4553 	} else if (procnum == NFSV4OP_CBRECALL) {
4554 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4555 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4556 		    "CB Recall", &sep, &slotpos);
4557 		if (error != 0) {
4558 			m_freem(nd->nd_mreq);
4559 			goto errout;
4560 		}
4561 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4562 		*tl++ = txdr_unsigned(stateidp->seqid);
4563 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4564 		    NFSX_STATEIDOTHER);
4565 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4566 		if (trunc)
4567 			*tl = newnfs_true;
4568 		else
4569 			*tl = newnfs_false;
4570 		(void)nfsm_fhtom(NULL, nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4571 	} else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4572 		NFSD_DEBUG(4, "docallback layout recall\n");
4573 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4574 		error = nfsrv_cbcallargs(nd, clp, callback,
4575 		    NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep, &slotpos);
4576 		NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4577 		if (error != 0) {
4578 			m_freem(nd->nd_mreq);
4579 			goto errout;
4580 		}
4581 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4582 		*tl++ = txdr_unsigned(laytype);
4583 		*tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4584 		if (trunc)
4585 			*tl++ = newnfs_true;
4586 		else
4587 			*tl++ = newnfs_false;
4588 		*tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4589 		(void)nfsm_fhtom(NULL, nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4590 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4591 		tval = 0;
4592 		txdr_hyper(tval, tl); tl += 2;
4593 		tval = UINT64_MAX;
4594 		txdr_hyper(tval, tl); tl += 2;
4595 		*tl++ = txdr_unsigned(stateidp->seqid);
4596 		NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4597 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4598 		NFSD_DEBUG(4, "aft args\n");
4599 	} else if (procnum == NFSV4PROC_CBNULL) {
4600 		nd->nd_procnum = NFSV4PROC_CBNULL;
4601 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4602 			error = nfsv4_getcbsession(clp, &sep);
4603 			if (error != 0) {
4604 				m_freem(nd->nd_mreq);
4605 				goto errout;
4606 			}
4607 		}
4608 	} else {
4609 		error = NFSERR_SERVERFAULT;
4610 		m_freem(nd->nd_mreq);
4611 		goto errout;
4612 	}
4613 
4614 	/*
4615 	 * Call newnfs_connect(), as required, and then newnfs_request().
4616 	 */
4617 	dotls = false;
4618 	if ((clp->lc_flags & LCL_TLSCB) != 0)
4619 		dotls = true;
4620 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4621 	if (clp->lc_req.nr_client == NULL) {
4622 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4623 			error = ECONNREFUSED;
4624 			if (procnum != NFSV4PROC_CBNULL)
4625 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4626 				    true);
4627 			nfsrv_freesession(NULL, sep, NULL);
4628 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4629 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4630 			    NULL, 1, dotls, &clp->lc_req.nr_client);
4631 		else
4632 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4633 			    NULL, 3, dotls, &clp->lc_req.nr_client);
4634 	}
4635 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4636 	NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4637 	if (!error) {
4638 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4639 			KASSERT(sep != NULL, ("sep NULL"));
4640 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4641 				error = newnfs_request(nd, NULL, clp,
4642 				    &clp->lc_req, NULL, NULL, cred,
4643 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4644 				    1, NULL, &sep->sess_cbsess);
4645 			else {
4646 				/*
4647 				 * This should probably never occur, but if a
4648 				 * client somehow does an RPC without a
4649 				 * SequenceID Op that causes a callback just
4650 				 * after the nfsd threads have been terminated
4651 				 * and restarted we could conceivably get here
4652 				 * without a backchannel xprt.
4653 				 */
4654 				printf("nfsrv_docallback: no xprt\n");
4655 				error = ECONNREFUSED;
4656 			}
4657 			NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4658 			if (error != 0 && procnum != NFSV4PROC_CBNULL) {
4659 				/*
4660 				 * It is likely that the callback was never
4661 				 * processed by the client and, as such,
4662 				 * the sequence# for the session slot needs
4663 				 * to be backed up by one to avoid a
4664 				 * NFSERR_SEQMISORDERED error reply.
4665 				 * For the unlikely case where the callback
4666 				 * was processed by the client, this will
4667 				 * make the next callback on the slot
4668 				 * appear to be a retry.
4669 				 * Since callbacks never specify that the
4670 				 * reply be cached, this "apparent retry"
4671 				 * should not be a problem.
4672 				 */
4673 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4674 				    true);
4675 			}
4676 			nfsrv_freesession(NULL, sep, NULL);
4677 		} else
4678 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4679 			    NULL, NULL, cred, clp->lc_program,
4680 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4681 	}
4682 errout:
4683 	NFSFREECRED(cred);
4684 
4685 	/*
4686 	 * If error is set here, the Callback path isn't working
4687 	 * properly, so twiddle the appropriate LCL_ flags.
4688 	 * (nd_repstat != 0 indicates the Callback path is working,
4689 	 *  but the callback failed on the client.)
4690 	 */
4691 	if (error) {
4692 		/*
4693 		 * Mark the callback pathway down, which disabled issuing
4694 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4695 		 */
4696 		NFSLOCKSTATE();
4697 		clp->lc_flags |= LCL_CBDOWN;
4698 		NFSUNLOCKSTATE();
4699 	} else {
4700 		/*
4701 		 * Callback worked. If the callback path was down, disable
4702 		 * callbacks, so no more delegations will be issued. (This
4703 		 * is done on the assumption that the callback pathway is
4704 		 * flakey.)
4705 		 */
4706 		NFSLOCKSTATE();
4707 		if (clp->lc_flags & LCL_CBDOWN)
4708 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4709 		NFSUNLOCKSTATE();
4710 		if (nd->nd_repstat) {
4711 			error = nd->nd_repstat;
4712 			NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4713 			    procnum, error);
4714 		} else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4715 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4716 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4717 			    p, NULL);
4718 		m_freem(nd->nd_mrep);
4719 	}
4720 	NFSLOCKSTATE();
4721 	clp->lc_cbref--;
4722 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4723 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4724 		wakeup(clp);
4725 	}
4726 	NFSUNLOCKSTATE();
4727 
4728 	free(nd, M_TEMP);
4729 	NFSEXITCODE(error);
4730 	return (error);
4731 }
4732 
4733 /*
4734  * Set up the compound RPC for the callback.
4735  */
4736 static int
4737 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4738     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
4739     int *slotposp)
4740 {
4741 	uint32_t *tl;
4742 	int error, len;
4743 
4744 	len = strlen(optag);
4745 	(void)nfsm_strtom(nd, optag, len);
4746 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4747 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4748 		if ((nd->nd_flag & ND_NFSV42) != 0)
4749 			*tl++ = txdr_unsigned(NFSV42_MINORVERSION);
4750 		else
4751 			*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4752 		*tl++ = txdr_unsigned(callback);
4753 		*tl++ = txdr_unsigned(2);
4754 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4755 		error = nfsv4_setcbsequence(nd, clp, 1, sepp, slotposp);
4756 		if (error != 0)
4757 			return (error);
4758 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4759 		*tl = txdr_unsigned(op);
4760 	} else {
4761 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4762 		*tl++ = txdr_unsigned(callback);
4763 		*tl++ = txdr_unsigned(1);
4764 		*tl = txdr_unsigned(op);
4765 	}
4766 	return (0);
4767 }
4768 
4769 /*
4770  * Return the next index# for a clientid. Mostly just increment and return
4771  * the next one, but... if the 32bit unsigned does actually wrap around,
4772  * it should be rebooted.
4773  * At an average rate of one new client per second, it will wrap around in
4774  * approximately 136 years. (I think the server will have been shut
4775  * down or rebooted before then.)
4776  */
4777 static u_int32_t
4778 nfsrv_nextclientindex(void)
4779 {
4780 	static u_int32_t client_index = 0;
4781 
4782 	client_index++;
4783 	if (client_index != 0)
4784 		return (client_index);
4785 
4786 	printf("%s: out of clientids\n", __func__);
4787 	return (client_index);
4788 }
4789 
4790 /*
4791  * Return the next index# for a stateid. Mostly just increment and return
4792  * the next one, but... if the 32bit unsigned does actually wrap around
4793  * (will a BSD server stay up that long?), find
4794  * new start and end values.
4795  */
4796 static u_int32_t
4797 nfsrv_nextstateindex(struct nfsclient *clp)
4798 {
4799 	struct nfsstate *stp;
4800 	int i;
4801 	u_int32_t canuse, min_index, max_index;
4802 
4803 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4804 		clp->lc_stateindex++;
4805 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4806 			return (clp->lc_stateindex);
4807 	}
4808 
4809 	/*
4810 	 * Yuck, we've hit the end.
4811 	 * Look for a new min and max.
4812 	 */
4813 	min_index = 0;
4814 	max_index = 0xffffffff;
4815 	for (i = 0; i < nfsrv_statehashsize; i++) {
4816 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4817 		if (stp->ls_stateid.other[2] > 0x80000000) {
4818 		    if (stp->ls_stateid.other[2] < max_index)
4819 			max_index = stp->ls_stateid.other[2];
4820 		} else {
4821 		    if (stp->ls_stateid.other[2] > min_index)
4822 			min_index = stp->ls_stateid.other[2];
4823 		}
4824 	    }
4825 	}
4826 
4827 	/*
4828 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4829 	 */
4830 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4831 	    canuse = 0;
4832 	    /*
4833 	     * Loop around until we find an unused entry. Return that
4834 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4835 	     * (This is one of those rare cases where a goto is the
4836 	     *  cleanest way to code the loop.)
4837 	     */
4838 tryagain:
4839 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4840 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4841 		    if (stp->ls_stateid.other[2] == canuse) {
4842 			canuse++;
4843 			goto tryagain;
4844 		    }
4845 		}
4846 	    }
4847 	    clp->lc_flags |= LCL_INDEXNOTOK;
4848 	    return (canuse);
4849 	}
4850 
4851 	/*
4852 	 * Ok to start again from min + 1.
4853 	 */
4854 	clp->lc_stateindex = min_index + 1;
4855 	clp->lc_statemaxindex = max_index;
4856 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4857 	return (clp->lc_stateindex);
4858 }
4859 
4860 /*
4861  * The following functions handle the stable storage file that deals with
4862  * the edge conditions described in RFC3530 Sec. 8.6.3.
4863  * The file is as follows:
4864  * - a single record at the beginning that has the lease time of the
4865  *   previous server instance (before the last reboot) and the nfsrvboottime
4866  *   values for the previous server boots.
4867  *   These previous boot times are used to ensure that the current
4868  *   nfsrvboottime does not, somehow, get set to a previous one.
4869  *   (This is important so that Stale ClientIDs and StateIDs can
4870  *    be recognized.)
4871  *   The number of previous nfsvrboottime values precedes the list.
4872  * - followed by some number of appended records with:
4873  *   - client id string
4874  *   - flag that indicates it is a record revoking state via lease
4875  *     expiration or similar
4876  *     OR has successfully acquired state.
4877  * These structures vary in length, with the client string at the end, up
4878  * to NFSV4_OPAQUELIMIT in size.
4879  *
4880  * At the end of the grace period, the file is truncated, the first
4881  * record is rewritten with updated information and any acquired state
4882  * records for successful reclaims of state are written.
4883  *
4884  * Subsequent records are appended when the first state is issued to
4885  * a client and when state is revoked for a client.
4886  *
4887  * When reading the file in, state issued records that come later in
4888  * the file override older ones, since the append log is in cronological order.
4889  * If, for some reason, the file can't be read, the grace period is
4890  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4891  */
4892 
4893 /*
4894  * Read in the stable storage file. Called by nfssvc() before the nfsd
4895  * processes start servicing requests.
4896  */
4897 void
4898 nfsrv_setupstable(NFSPROC_T *p)
4899 {
4900 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
4901 	struct nfsrv_stable *sp, *nsp;
4902 	struct nfst_rec *tsp;
4903 	int error, i, tryagain;
4904 	off_t off = 0;
4905 	ssize_t aresid, len;
4906 
4907 	/*
4908 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4909 	 * a reboot, so state has not been lost.
4910 	 */
4911 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4912 		return;
4913 	/*
4914 	 * Set Grace over just until the file reads successfully.
4915 	 */
4916 	NFSD_VNET(nfsrvboottime) = time_second;
4917 	LIST_INIT(&sf->nsf_head);
4918 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4919 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4920 	if (sf->nsf_fp == NULL)
4921 		return;
4922 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4923 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4924 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4925 	if (error || aresid || sf->nsf_numboots == 0 ||
4926 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4927 		return;
4928 
4929 	/*
4930 	 * Now, read in the boottimes.
4931 	 */
4932 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4933 		sizeof(time_t), M_TEMP, M_WAITOK);
4934 	off = sizeof (struct nfsf_rec);
4935 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4936 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4937 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4938 	if (error || aresid) {
4939 		free(sf->nsf_bootvals, M_TEMP);
4940 		sf->nsf_bootvals = NULL;
4941 		return;
4942 	}
4943 
4944 	/*
4945 	 * Make sure this nfsrvboottime is different from all recorded
4946 	 * previous ones.
4947 	 */
4948 	do {
4949 		tryagain = 0;
4950 		for (i = 0; i < sf->nsf_numboots; i++) {
4951 			if (NFSD_VNET(nfsrvboottime) == sf->nsf_bootvals[i]) {
4952 				NFSD_VNET(nfsrvboottime)++;
4953 				tryagain = 1;
4954 				break;
4955 			}
4956 		}
4957 	} while (tryagain);
4958 
4959 	sf->nsf_flags |= NFSNSF_OK;
4960 	off += (sf->nsf_numboots * sizeof (time_t));
4961 
4962 	/*
4963 	 * Read through the file, building a list of records for grace
4964 	 * checking.
4965 	 * Each record is between sizeof (struct nfst_rec) and
4966 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4967 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4968 	 */
4969 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4970 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4971 	do {
4972 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4973 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4974 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4975 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4976 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4977 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4978 		/*
4979 		 * Yuck, the file has been corrupted, so just return
4980 		 * after clearing out any restart state, so the grace period
4981 		 * is over.
4982 		 */
4983 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4984 			LIST_REMOVE(sp, nst_list);
4985 			free(sp, M_TEMP);
4986 		}
4987 		free(tsp, M_TEMP);
4988 		sf->nsf_flags &= ~NFSNSF_OK;
4989 		free(sf->nsf_bootvals, M_TEMP);
4990 		sf->nsf_bootvals = NULL;
4991 		return;
4992 	    }
4993 	    if (len > 0) {
4994 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4995 		/*
4996 		 * Search the list for a matching client.
4997 		 */
4998 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4999 			if (tsp->len == sp->nst_len &&
5000 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
5001 				break;
5002 		}
5003 		if (sp == LIST_END(&sf->nsf_head)) {
5004 			sp = (struct nfsrv_stable *)malloc(tsp->len +
5005 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
5006 				M_WAITOK);
5007 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
5008 				sizeof (struct nfst_rec) + tsp->len - 1);
5009 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
5010 		} else {
5011 			if (tsp->flag == NFSNST_REVOKE)
5012 				sp->nst_flag |= NFSNST_REVOKE;
5013 			else
5014 				/*
5015 				 * A subsequent timestamp indicates the client
5016 				 * did a setclientid/confirm and any previous
5017 				 * revoke is no longer relevant.
5018 				 */
5019 				sp->nst_flag &= ~NFSNST_REVOKE;
5020 		}
5021 	    }
5022 	} while (len > 0);
5023 	free(tsp, M_TEMP);
5024 	sf->nsf_flags = NFSNSF_OK;
5025 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
5026 		NFSRV_LEASEDELTA;
5027 }
5028 
5029 /*
5030  * Update the stable storage file, now that the grace period is over.
5031  */
5032 void
5033 nfsrv_updatestable(NFSPROC_T *p)
5034 {
5035 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
5036 	struct nfsrv_stable *sp, *nsp;
5037 	int i;
5038 	struct nfsvattr nva;
5039 	vnode_t vp;
5040 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
5041 	mount_t mp = NULL;
5042 #endif
5043 	int error;
5044 
5045 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
5046 		return;
5047 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
5048 	/*
5049 	 * Ok, we need to rewrite the stable storage file.
5050 	 * - truncate to 0 length
5051 	 * - write the new first structure
5052 	 * - loop through the data structures, writing out any that
5053 	 *   have timestamps older than the old boot
5054 	 */
5055 	if (sf->nsf_bootvals) {
5056 		sf->nsf_numboots++;
5057 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
5058 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
5059 	} else {
5060 		sf->nsf_numboots = 1;
5061 		sf->nsf_bootvals = (time_t *)malloc(sizeof(time_t),
5062 			M_TEMP, M_WAITOK);
5063 	}
5064 	sf->nsf_bootvals[0] = NFSD_VNET(nfsrvboottime);
5065 	sf->nsf_lease = nfsrv_lease;
5066 	NFSVNO_ATTRINIT(&nva);
5067 	NFSVNO_SETATTRVAL(&nva, size, 0);
5068 	vp = NFSFPVNODE(sf->nsf_fp);
5069 	vn_start_write(vp, &mp, V_WAIT);
5070 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5071 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
5072 		    NULL);
5073 		NFSVOPUNLOCK(vp);
5074 	} else
5075 		error = EPERM;
5076 	vn_finished_write(mp);
5077 	if (!error)
5078 	    error = NFSD_RDWR(UIO_WRITE, vp,
5079 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
5080 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5081 	if (!error)
5082 	    error = NFSD_RDWR(UIO_WRITE, vp,
5083 		(caddr_t)sf->nsf_bootvals,
5084 		sf->nsf_numboots * sizeof (time_t),
5085 		(off_t)(sizeof (struct nfsf_rec)),
5086 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5087 	free(sf->nsf_bootvals, M_TEMP);
5088 	sf->nsf_bootvals = NULL;
5089 	if (error) {
5090 		sf->nsf_flags &= ~NFSNSF_OK;
5091 		printf("EEK! Can't write NfsV4 stable storage file\n");
5092 		return;
5093 	}
5094 	sf->nsf_flags |= NFSNSF_OK;
5095 
5096 	/*
5097 	 * Loop through the list and write out timestamp records for
5098 	 * any clients that successfully reclaimed state.
5099 	 */
5100 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
5101 		if (sp->nst_flag & NFSNST_GOTSTATE) {
5102 			nfsrv_writestable(sp->nst_client, sp->nst_len,
5103 				NFSNST_NEWSTATE, p);
5104 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
5105 		}
5106 		LIST_REMOVE(sp, nst_list);
5107 		free(sp, M_TEMP);
5108 	}
5109 	nfsrv_backupstable();
5110 }
5111 
5112 /*
5113  * Append a record to the stable storage file.
5114  */
5115 void
5116 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
5117 {
5118 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
5119 	struct nfst_rec *sp;
5120 	int error;
5121 
5122 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
5123 		return;
5124 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
5125 		len - 1, M_TEMP, M_WAITOK);
5126 	sp->len = len;
5127 	NFSBCOPY(client, sp->client, len);
5128 	sp->flag = flag;
5129 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
5130 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
5131 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
5132 	free(sp, M_TEMP);
5133 	if (error) {
5134 		sf->nsf_flags &= ~NFSNSF_OK;
5135 		printf("EEK! Can't write NfsV4 stable storage file\n");
5136 	}
5137 }
5138 
5139 /*
5140  * This function is called during the grace period to mark a client
5141  * that successfully reclaimed state.
5142  */
5143 static void
5144 nfsrv_markstable(struct nfsclient *clp)
5145 {
5146 	struct nfsrv_stable *sp;
5147 
5148 	/*
5149 	 * First find the client structure.
5150 	 */
5151 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5152 		if (sp->nst_len == clp->lc_idlen &&
5153 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5154 			break;
5155 	}
5156 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head))
5157 		return;
5158 
5159 	/*
5160 	 * Now, just mark it and set the nfsclient back pointer.
5161 	 */
5162 	sp->nst_flag |= NFSNST_GOTSTATE;
5163 	sp->nst_clp = clp;
5164 }
5165 
5166 /*
5167  * This function is called when a NFSv4.1 client does a ReclaimComplete.
5168  * Very similar to nfsrv_markstable(), except for the flag being set.
5169  */
5170 static void
5171 nfsrv_markreclaim(struct nfsclient *clp)
5172 {
5173 	struct nfsrv_stable *sp;
5174 
5175 	/*
5176 	 * First find the client structure.
5177 	 */
5178 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5179 		if (sp->nst_len == clp->lc_idlen &&
5180 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5181 			break;
5182 	}
5183 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head))
5184 		return;
5185 
5186 	/*
5187 	 * Now, just set the flag.
5188 	 */
5189 	sp->nst_flag |= NFSNST_RECLAIMED;
5190 }
5191 
5192 /*
5193  * This function is called for a reclaim, to see if it gets grace.
5194  * It returns 0 if a reclaim is allowed, 1 otherwise.
5195  */
5196 static int
5197 nfsrv_checkstable(struct nfsclient *clp)
5198 {
5199 	struct nfsrv_stable *sp;
5200 
5201 	/*
5202 	 * First, find the entry for the client.
5203 	 */
5204 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5205 		if (sp->nst_len == clp->lc_idlen &&
5206 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5207 			break;
5208 	}
5209 
5210 	/*
5211 	 * If not in the list, state was revoked or no state was issued
5212 	 * since the previous reboot, a reclaim is denied.
5213 	 */
5214 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head) ||
5215 	    (sp->nst_flag & NFSNST_REVOKE) ||
5216 	    !(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_OK))
5217 		return (1);
5218 	return (0);
5219 }
5220 
5221 /*
5222  * Test for and try to clear out a conflicting client. This is called by
5223  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5224  * a found.
5225  * The trick here is that it can't revoke a conflicting client with an
5226  * expired lease unless it holds the v4root lock, so...
5227  * If no v4root lock, get the lock and return 1 to indicate "try again".
5228  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5229  * the revocation worked and the conflicting client is "bye, bye", so it
5230  * can be tried again.
5231  * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK().
5232  * Unlocks State before a non-zero value is returned.
5233  */
5234 static int
5235 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5236     NFSPROC_T *p)
5237 {
5238 	int gotlock, lktype = 0;
5239 
5240 	/*
5241 	 * If lease hasn't expired, we can't fix it.
5242 	 */
5243 	if (clp->lc_expiry >= NFSD_MONOSEC ||
5244 	    !(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE))
5245 		return (0);
5246 	if (*haslockp == 0) {
5247 		NFSUNLOCKSTATE();
5248 		if (vp != NULL) {
5249 			lktype = NFSVOPISLOCKED(vp);
5250 			NFSVOPUNLOCK(vp);
5251 		}
5252 		NFSLOCKV4ROOTMUTEX();
5253 		nfsv4_relref(&nfsv4rootfs_lock);
5254 		do {
5255 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5256 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5257 		} while (!gotlock);
5258 		NFSUNLOCKV4ROOTMUTEX();
5259 		*haslockp = 1;
5260 		if (vp != NULL) {
5261 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5262 			if (VN_IS_DOOMED(vp))
5263 				return (2);
5264 		}
5265 		return (1);
5266 	}
5267 	NFSUNLOCKSTATE();
5268 
5269 	/*
5270 	 * Ok, we can expire the conflicting client.
5271 	 */
5272 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5273 	nfsrv_backupstable();
5274 	nfsrv_cleanclient(clp, p);
5275 	nfsrv_freedeleglist(&clp->lc_deleg);
5276 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5277 	LIST_REMOVE(clp, lc_hash);
5278 	nfsrv_zapclient(clp, p);
5279 	return (1);
5280 }
5281 
5282 /*
5283  * Resolve a delegation conflict.
5284  * Returns 0 to indicate the conflict was resolved without sleeping.
5285  * Return -1 to indicate that the caller should check for conflicts again.
5286  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5287  *
5288  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5289  * for a return of 0, since there was no sleep and it could be required
5290  * later. It is released for a return of NFSERR_DELAY, since the caller
5291  * will return that error. It is released when a sleep was done waiting
5292  * for the delegation to be returned or expire (so that other nfsds can
5293  * handle ops). Then, it must be acquired for the write to stable storage.
5294  * (This function is somewhat similar to nfsrv_clientconflict(), but
5295  *  the semantics differ in a couple of subtle ways. The return of 0
5296  *  indicates the conflict was resolved without sleeping here, not
5297  *  that the conflict can't be resolved and the handling of nfsv4root_lock
5298  *  differs, as noted above.)
5299  * Unlocks State before returning a non-zero value.
5300  */
5301 static int
5302 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5303     vnode_t vp)
5304 {
5305 	struct nfsclient *clp = stp->ls_clp;
5306 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5307 	nfsv4stateid_t tstateid;
5308 	fhandle_t tfh;
5309 
5310 	/*
5311 	 * If the conflict is with an old delegation...
5312 	 */
5313 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
5314 		/*
5315 		 * You can delete it, if it has expired.
5316 		 */
5317 		if (clp->lc_delegtime < NFSD_MONOSEC) {
5318 			nfsrv_freedeleg(stp);
5319 			NFSUNLOCKSTATE();
5320 			error = -1;
5321 			goto out;
5322 		}
5323 		NFSUNLOCKSTATE();
5324 		/*
5325 		 * During this delay, the old delegation could expire or it
5326 		 * could be recovered by the client via an Open with
5327 		 * CLAIM_DELEGATE_PREV.
5328 		 * Release the nfsv4root_lock, if held.
5329 		 */
5330 		if (*haslockp) {
5331 			*haslockp = 0;
5332 			NFSLOCKV4ROOTMUTEX();
5333 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5334 			NFSUNLOCKV4ROOTMUTEX();
5335 		}
5336 		error = NFSERR_DELAY;
5337 		goto out;
5338 	}
5339 
5340 	/*
5341 	 * It's a current delegation, so:
5342 	 * - check to see if the delegation has expired
5343 	 *   - if so, get the v4root lock and then expire it
5344 	 */
5345 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0 || (stp->ls_lastrecall <
5346 	    NFSD_MONOSEC && clp->lc_expiry >= NFSD_MONOSEC &&
5347 	    stp->ls_delegtime >= NFSD_MONOSEC)) {
5348 		/*
5349 		 * - do a recall callback, since not yet done
5350 		 * For now, never allow truncate to be set. To use
5351 		 * truncate safely, it must be guaranteed that the
5352 		 * Remove, Rename or Setattr with size of 0 will
5353 		 * succeed and that would require major changes to
5354 		 * the VFS/Vnode OPs.
5355 		 * Set the expiry time large enough so that it won't expire
5356 		 * until after the callback, then set it correctly, once
5357 		 * the callback is done. (The delegation will now time
5358 		 * out whether or not the Recall worked ok. The timeout
5359 		 * will be extended when ops are done on the delegation
5360 		 * stateid, up to the timelimit.)
5361 		 */
5362 		if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) {
5363 			stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5364 			    NFSRV_LEASEDELTA;
5365 			stp->ls_delegtimelimit = NFSD_MONOSEC + (6 *
5366 			    nfsrv_lease) + NFSRV_LEASEDELTA;
5367 			stp->ls_flags |= NFSLCK_DELEGRECALL;
5368 		}
5369 		stp->ls_lastrecall = time_uptime + 1;
5370 
5371 		/*
5372 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5373 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5374 		 * in order to try and avoid a race that could happen
5375 		 * when a CBRecall request passed the Open reply with
5376 		 * the delegation in it when transitting the network.
5377 		 * Since nfsrv_docallback will sleep, don't use stp after
5378 		 * the call.
5379 		 */
5380 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5381 		    sizeof (tstateid));
5382 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5383 		    sizeof (tfh));
5384 		NFSUNLOCKSTATE();
5385 		if (*haslockp) {
5386 			*haslockp = 0;
5387 			NFSLOCKV4ROOTMUTEX();
5388 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5389 			NFSUNLOCKV4ROOTMUTEX();
5390 		}
5391 		retrycnt = 0;
5392 		do {
5393 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5394 			&tstateid, 0, &tfh, NULL, NULL, 0, p);
5395 		    retrycnt++;
5396 		} while ((error == NFSERR_BADSTATEID ||
5397 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5398 		error = NFSERR_DELAY;
5399 		goto out;
5400 	}
5401 
5402 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5403 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5404 		NFSUNLOCKSTATE();
5405 		/*
5406 		 * A recall has been done, but it has not yet expired.
5407 		 * So, RETURN_DELAY.
5408 		 */
5409 		if (*haslockp) {
5410 			*haslockp = 0;
5411 			NFSLOCKV4ROOTMUTEX();
5412 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5413 			NFSUNLOCKV4ROOTMUTEX();
5414 		}
5415 		error = NFSERR_DELAY;
5416 		goto out;
5417 	}
5418 
5419 	/*
5420 	 * If we don't yet have the lock, just get it and then return,
5421 	 * since we need that before deleting expired state, such as
5422 	 * this delegation.
5423 	 * When getting the lock, unlock the vnode, so other nfsds that
5424 	 * are in progress, won't get stuck waiting for the vnode lock.
5425 	 */
5426 	if (*haslockp == 0) {
5427 		NFSUNLOCKSTATE();
5428 		if (vp != NULL) {
5429 			lktype = NFSVOPISLOCKED(vp);
5430 			NFSVOPUNLOCK(vp);
5431 		}
5432 		NFSLOCKV4ROOTMUTEX();
5433 		nfsv4_relref(&nfsv4rootfs_lock);
5434 		do {
5435 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5436 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5437 		} while (!gotlock);
5438 		NFSUNLOCKV4ROOTMUTEX();
5439 		*haslockp = 1;
5440 		if (vp != NULL) {
5441 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5442 			if (VN_IS_DOOMED(vp)) {
5443 				*haslockp = 0;
5444 				NFSLOCKV4ROOTMUTEX();
5445 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5446 				NFSUNLOCKV4ROOTMUTEX();
5447 				error = NFSERR_PERM;
5448 				goto out;
5449 			}
5450 		}
5451 		error = -1;
5452 		goto out;
5453 	}
5454 
5455 	NFSUNLOCKSTATE();
5456 	/*
5457 	 * Ok, we can delete the expired delegation.
5458 	 * First, write the Revoke record to stable storage and then
5459 	 * clear out the conflict.
5460 	 * Since all other nfsd threads are now blocked, we can safely
5461 	 * sleep without the state changing.
5462 	 */
5463 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5464 	nfsrv_backupstable();
5465 	if (clp->lc_expiry < NFSD_MONOSEC) {
5466 		nfsrv_cleanclient(clp, p);
5467 		nfsrv_freedeleglist(&clp->lc_deleg);
5468 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5469 		LIST_REMOVE(clp, lc_hash);
5470 		zapped_clp = 1;
5471 	} else {
5472 		nfsrv_freedeleg(stp);
5473 		zapped_clp = 0;
5474 	}
5475 	if (zapped_clp)
5476 		nfsrv_zapclient(clp, p);
5477 	error = -1;
5478 
5479 out:
5480 	NFSEXITCODE(error);
5481 	return (error);
5482 }
5483 
5484 /*
5485  * Check for a remove allowed, if remove is set to 1 and get rid of
5486  * delegations.
5487  */
5488 int
5489 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd,
5490     nfsquad_t clientid, NFSPROC_T *p)
5491 {
5492 	struct nfsclient *clp;
5493 	struct nfsstate *stp;
5494 	struct nfslockfile *lfp;
5495 	int error, haslock = 0;
5496 	fhandle_t nfh;
5497 
5498 	clp = NULL;
5499 	/*
5500 	 * First, get the lock file structure.
5501 	 * (A return of -1 means no associated state, so remove ok.)
5502 	 */
5503 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5504 tryagain:
5505 	NFSLOCKSTATE();
5506 	if (error == 0 && clientid.qval != 0)
5507 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
5508 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
5509 	if (!error)
5510 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5511 	if (error) {
5512 		NFSUNLOCKSTATE();
5513 		if (haslock) {
5514 			NFSLOCKV4ROOTMUTEX();
5515 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5516 			NFSUNLOCKV4ROOTMUTEX();
5517 		}
5518 		if (error == -1)
5519 			error = 0;
5520 		goto out;
5521 	}
5522 
5523 	/*
5524 	 * Now, we must Recall any delegations.
5525 	 */
5526 	error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
5527 	if (error) {
5528 		/*
5529 		 * nfsrv_cleandeleg() unlocks state for non-zero
5530 		 * return.
5531 		 */
5532 		if (error == -1)
5533 			goto tryagain;
5534 		if (haslock) {
5535 			NFSLOCKV4ROOTMUTEX();
5536 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5537 			NFSUNLOCKV4ROOTMUTEX();
5538 		}
5539 		goto out;
5540 	}
5541 
5542 	/*
5543 	 * Now, look for a conflicting open share.
5544 	 */
5545 	if (remove) {
5546 		/*
5547 		 * If the entry in the directory was the last reference to the
5548 		 * corresponding filesystem object, the object can be destroyed
5549 		 * */
5550 		if(lfp->lf_usecount>1)
5551 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5552 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5553 					error = NFSERR_FILEOPEN;
5554 					break;
5555 				}
5556 			}
5557 	}
5558 
5559 	NFSUNLOCKSTATE();
5560 	if (haslock) {
5561 		NFSLOCKV4ROOTMUTEX();
5562 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5563 		NFSUNLOCKV4ROOTMUTEX();
5564 	}
5565 
5566 out:
5567 	NFSEXITCODE(error);
5568 	return (error);
5569 }
5570 
5571 /*
5572  * Clear out all delegations for the file referred to by lfp.
5573  * May return NFSERR_DELAY, if there will be a delay waiting for
5574  * delegations to expire.
5575  * Returns -1 to indicate it slept while recalling a delegation.
5576  * This function has the side effect of deleting the nfslockfile structure,
5577  * if it no longer has associated state and didn't have to sleep.
5578  * Unlocks State before a non-zero value is returned.
5579  */
5580 static int
5581 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5582     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5583 {
5584 	struct nfsstate *stp, *nstp;
5585 	int ret = 0;
5586 
5587 	stp = LIST_FIRST(&lfp->lf_deleg);
5588 	while (stp != LIST_END(&lfp->lf_deleg)) {
5589 		nstp = LIST_NEXT(stp, ls_file);
5590 		if (stp->ls_clp != clp) {
5591 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5592 			if (ret) {
5593 				/*
5594 				 * nfsrv_delegconflict() unlocks state
5595 				 * when it returns non-zero.
5596 				 */
5597 				goto out;
5598 			}
5599 		}
5600 		stp = nstp;
5601 	}
5602 out:
5603 	NFSEXITCODE(ret);
5604 	return (ret);
5605 }
5606 
5607 /*
5608  * There are certain operations that, when being done outside of NFSv4,
5609  * require that any NFSv4 delegation for the file be recalled.
5610  * This function is to be called for those cases:
5611  * VOP_RENAME() - When a delegation is being recalled for any reason,
5612  *	the client may have to do Opens against the server, using the file's
5613  *	final component name. If the file has been renamed on the server,
5614  *	that component name will be incorrect and the Open will fail.
5615  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5616  *	been removed on the server, if there is a delegation issued to
5617  *	that client for the file. I say "theoretically" since clients
5618  *	normally do an Access Op before the Open and that Access Op will
5619  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5620  *	they will detect the file's removal in the same manner. (There is
5621  *	one case where RFC3530 allows a client to do an Open without first
5622  *	doing an Access Op, which is passage of a check against the ACE
5623  *	returned with a Write delegation, but current practice is to ignore
5624  *	the ACE and always do an Access Op.)
5625  *	Since the functions can only be called with an unlocked vnode, this
5626  *	can't be done at this time.
5627  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5628  *	locks locally in the client, which are not visible to the server. To
5629  *	deal with this, issuing of delegations for a vnode must be disabled
5630  *	and all delegations for the vnode recalled. This is done via the
5631  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5632  */
5633 void
5634 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5635 {
5636 	time_t starttime;
5637 	int error;
5638 
5639 	/*
5640 	 * First, check to see if the server is currently running and it has
5641 	 * been called for a regular file when issuing delegations.
5642 	 */
5643 	if (NFSD_VNET(nfsrv_numnfsd) == 0 || vp->v_type != VREG ||
5644 	    nfsrv_issuedelegs == 0)
5645 		return;
5646 
5647 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5648 	/*
5649 	 * First, get a reference on the nfsv4rootfs_lock so that an
5650 	 * exclusive lock cannot be acquired by another thread.
5651 	 */
5652 	NFSLOCKV4ROOTMUTEX();
5653 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5654 	NFSUNLOCKV4ROOTMUTEX();
5655 
5656 	/*
5657 	 * Now, call nfsrv_checkremove() in a loop while it returns
5658 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5659 	 */
5660 	starttime = NFSD_MONOSEC;
5661 	do {
5662 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5663 			error = nfsrv_checkremove(vp, 0, NULL,
5664 			    (nfsquad_t)((u_quad_t)0), p);
5665 			NFSVOPUNLOCK(vp);
5666 		} else
5667 			error = EPERM;
5668 		if (error == NFSERR_DELAY) {
5669 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5670 				break;
5671 			/* Sleep for a short period of time */
5672 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5673 		}
5674 	} while (error == NFSERR_DELAY);
5675 	NFSLOCKV4ROOTMUTEX();
5676 	nfsv4_relref(&nfsv4rootfs_lock);
5677 	NFSUNLOCKV4ROOTMUTEX();
5678 }
5679 
5680 void
5681 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5682 {
5683 
5684 #ifdef VV_DISABLEDELEG
5685 	/*
5686 	 * First, flag issuance of delegations disabled.
5687 	 */
5688 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5689 #endif
5690 
5691 	/*
5692 	 * Then call nfsd_recalldelegation() to get rid of all extant
5693 	 * delegations.
5694 	 */
5695 	nfsd_recalldelegation(vp, p);
5696 }
5697 
5698 /*
5699  * Check for conflicting locks, etc. and then get rid of delegations.
5700  * (At one point I thought that I should get rid of delegations for any
5701  *  Setattr, since it could potentially disallow the I/O op (read or write)
5702  *  allowed by the delegation. However, Setattr Ops that aren't changing
5703  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5704  *  for the same client or a different one, so I decided to only get rid
5705  *  of delegations for other clients when the size is being changed.)
5706  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5707  * as Write backs, even if there is no delegation, so it really isn't any
5708  * different?)
5709  */
5710 int
5711 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5712     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5713     struct nfsexstuff *exp, NFSPROC_T *p)
5714 {
5715 	struct nfsstate st, *stp = &st;
5716 	struct nfslock lo, *lop = &lo;
5717 	int error = 0;
5718 	nfsquad_t clientid;
5719 
5720 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5721 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5722 		lop->lo_first = nvap->na_size;
5723 	} else {
5724 		stp->ls_flags = 0;
5725 		lop->lo_first = 0;
5726 	}
5727 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5728 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5729 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5730 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5731 		stp->ls_flags |= NFSLCK_SETATTR;
5732 	if (stp->ls_flags == 0)
5733 		goto out;
5734 	lop->lo_end = NFS64BITSSET;
5735 	lop->lo_flags = NFSLCK_WRITE;
5736 	stp->ls_ownerlen = 0;
5737 	stp->ls_op = NULL;
5738 	stp->ls_uid = nd->nd_cred->cr_uid;
5739 	stp->ls_stateid.seqid = stateidp->seqid;
5740 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5741 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5742 	stp->ls_stateid.other[2] = stateidp->other[2];
5743 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5744 	    stateidp, exp, nd, p);
5745 
5746 out:
5747 	NFSEXITCODE2(error, nd);
5748 	return (error);
5749 }
5750 
5751 /*
5752  * Check for a write delegation and do a CBGETATTR if there is one, updating
5753  * the attributes, as required.
5754  * Should I return an error if I can't get the attributes? (For now, I'll
5755  * just return ok.
5756  */
5757 int
5758 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5759     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5760 {
5761 	struct nfsstate *stp;
5762 	struct nfslockfile *lfp;
5763 	struct nfsclient *clp;
5764 	struct nfsvattr nva;
5765 	fhandle_t nfh;
5766 	int error = 0;
5767 	nfsattrbit_t cbbits;
5768 	u_quad_t delegfilerev;
5769 
5770 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5771 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5772 		goto out;
5773 	if (nfsrv_writedelegcnt == 0)
5774 		goto out;
5775 
5776 	/*
5777 	 * Get the lock file structure.
5778 	 * (A return of -1 means no associated state, so return ok.)
5779 	 */
5780 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5781 	NFSLOCKSTATE();
5782 	if (!error)
5783 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5784 	if (error) {
5785 		NFSUNLOCKSTATE();
5786 		if (error == -1)
5787 			error = 0;
5788 		goto out;
5789 	}
5790 
5791 	/*
5792 	 * Now, look for a write delegation.
5793 	 */
5794 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5795 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5796 			break;
5797 	}
5798 	if (stp == LIST_END(&lfp->lf_deleg)) {
5799 		NFSUNLOCKSTATE();
5800 		goto out;
5801 	}
5802 	clp = stp->ls_clp;
5803 
5804 	/* If the clientid is not confirmed, ignore the delegation. */
5805 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
5806 		NFSUNLOCKSTATE();
5807 		goto out;
5808 	}
5809 
5810 	delegfilerev = stp->ls_filerev;
5811 	/*
5812 	 * If the Write delegation was issued as a part of this Compound RPC
5813 	 * or if we have an Implied Clientid (used in a previous Op in this
5814 	 * compound) and it is the client the delegation was issued to,
5815 	 * just return ok.
5816 	 * I also assume that it is from the same client iff the network
5817 	 * host IP address is the same as the callback address. (Not
5818 	 * exactly correct by the RFC, but avoids a lot of Getattr
5819 	 * callbacks.)
5820 	 */
5821 	if (nd->nd_compref == stp->ls_compref ||
5822 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5823 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5824 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5825 		NFSUNLOCKSTATE();
5826 		goto out;
5827 	}
5828 
5829 	/*
5830 	 * We are now done with the delegation state structure,
5831 	 * so the statelock can be released and we can now tsleep().
5832 	 */
5833 
5834 	/*
5835 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5836 	 * has changed.
5837 	 */
5838 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5839 		NFSUNLOCKSTATE();
5840 		NFSVNO_ATTRINIT(&nva);
5841 		nva.na_filerev = NFS64BITSSET;
5842 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5843 		    0, &nfh, &nva, &cbbits, 0, p);
5844 		if (!error) {
5845 			if ((nva.na_filerev != NFS64BITSSET &&
5846 			    nva.na_filerev > delegfilerev) ||
5847 			    (NFSVNO_ISSETSIZE(&nva) &&
5848 			     nva.na_size != nvap->na_size)) {
5849 				error = nfsvno_updfilerev(vp, nvap, nd, p);
5850 				if (NFSVNO_ISSETSIZE(&nva))
5851 					nvap->na_size = nva.na_size;
5852 			}
5853 		} else
5854 			error = 0;	/* Ignore callback errors for now. */
5855 	} else {
5856 		NFSUNLOCKSTATE();
5857 	}
5858 
5859 out:
5860 	NFSEXITCODE2(error, nd);
5861 	return (error);
5862 }
5863 
5864 /*
5865  * This function looks for openowners that haven't had any opens for
5866  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5867  * is set.
5868  */
5869 void
5870 nfsrv_throwawayopens(NFSPROC_T *p)
5871 {
5872 	struct nfsclient *clp, *nclp;
5873 	struct nfsstate *stp, *nstp;
5874 	int i;
5875 
5876 	NFSLOCKSTATE();
5877 	NFSD_VNET(nfsrv_stablefirst).nsf_flags &= ~NFSNSF_NOOPENS;
5878 	/*
5879 	 * For each client...
5880 	 */
5881 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5882 	    LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash,
5883 		nclp) {
5884 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5885 			if (LIST_EMPTY(&stp->ls_open) &&
5886 			    (stp->ls_noopens > NFSNOOPEN ||
5887 			     (nfsrv_openpluslock * 2) >
5888 			     nfsrv_v4statelimit))
5889 				nfsrv_freeopenowner(stp, 0, p);
5890 		}
5891 	    }
5892 	}
5893 	NFSUNLOCKSTATE();
5894 }
5895 
5896 /*
5897  * This function checks to see if the credentials are the same.
5898  * The check for same credentials is needed for state management operations
5899  * for NFSv4.0 or NFSv4.1/4.2 when SP4_MACH_CRED is configured via
5900  * ExchangeID.
5901  * Returns 1 for not same, 0 otherwise.
5902  */
5903 static int
5904 nfsrv_notsamecredname(int op, struct nfsrv_descript *nd, struct nfsclient *clp)
5905 {
5906 
5907 	/* Check for the SP4_MACH_CRED case. */
5908 	if (op != 0 && nfsrv_checkmachcred(op, nd, clp) != 0)
5909 		return (1);
5910 
5911 	/* For NFSv4.1/4.2, SP4_NONE always allows this. */
5912 	if ((nd->nd_flag & ND_NFSV41) != 0)
5913 		return (0);
5914 
5915 	if (nd->nd_flag & ND_GSS) {
5916 		if (!(clp->lc_flags & LCL_GSS))
5917 			return (1);
5918 		if (clp->lc_flags & LCL_NAME) {
5919 			if (nd->nd_princlen != clp->lc_namelen ||
5920 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5921 				clp->lc_namelen))
5922 				return (1);
5923 			else
5924 				return (0);
5925 		}
5926 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5927 			return (0);
5928 		else
5929 			return (1);
5930 	} else if (clp->lc_flags & LCL_GSS)
5931 		return (1);
5932 	/*
5933 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5934 	 * in RFC3530, which talks about principals, but doesn't say anything
5935 	 * about uids for AUTH_SYS.)
5936 	 */
5937 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5938 		return (0);
5939 	else
5940 		return (1);
5941 }
5942 
5943 /*
5944  * Calculate the lease expiry time.
5945  */
5946 static time_t
5947 nfsrv_leaseexpiry(void)
5948 {
5949 
5950 	if (NFSD_VNET(nfsrv_stablefirst).nsf_eograce > NFSD_MONOSEC)
5951 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5952 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5953 }
5954 
5955 /*
5956  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5957  */
5958 static void
5959 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5960 {
5961 
5962 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5963 		return;
5964 
5965 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5966 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5967 		stp->ls_delegtime += nfsrv_lease;
5968 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5969 			stp->ls_delegtime = stp->ls_delegtimelimit;
5970 	}
5971 }
5972 
5973 /*
5974  * This function checks to see if there is any other state associated
5975  * with the openowner for this Open.
5976  * It returns 1 if there is no other state, 0 otherwise.
5977  */
5978 static int
5979 nfsrv_nootherstate(struct nfsstate *stp)
5980 {
5981 	struct nfsstate *tstp;
5982 
5983 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5984 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5985 			return (0);
5986 	}
5987 	return (1);
5988 }
5989 
5990 /*
5991  * Create a list of lock deltas (changes to local byte range locking
5992  * that can be rolled back using the list) and apply the changes via
5993  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5994  * the rollback or update function will be called after this.
5995  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5996  * call fails. If it returns an error, it will unlock the list.
5997  */
5998 static int
5999 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
6000     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
6001 {
6002 	struct nfslock *lop, *nlop;
6003 	int error = 0;
6004 
6005 	/* Loop through the list of locks. */
6006 	lop = LIST_FIRST(&lfp->lf_locallock);
6007 	while (first < end && lop != NULL) {
6008 		nlop = LIST_NEXT(lop, lo_lckowner);
6009 		if (first >= lop->lo_end) {
6010 			/* not there yet */
6011 			lop = nlop;
6012 		} else if (first < lop->lo_first) {
6013 			/* new one starts before entry in list */
6014 			if (end <= lop->lo_first) {
6015 				/* no overlap between old and new */
6016 				error = nfsrv_dolocal(vp, lfp, flags,
6017 				    NFSLCK_UNLOCK, first, end, cfp, p);
6018 				if (error != 0)
6019 					break;
6020 				first = end;
6021 			} else {
6022 				/* handle fragment overlapped with new one */
6023 				error = nfsrv_dolocal(vp, lfp, flags,
6024 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
6025 				    p);
6026 				if (error != 0)
6027 					break;
6028 				first = lop->lo_first;
6029 			}
6030 		} else {
6031 			/* new one overlaps this entry in list */
6032 			if (end <= lop->lo_end) {
6033 				/* overlaps all of new one */
6034 				error = nfsrv_dolocal(vp, lfp, flags,
6035 				    lop->lo_flags, first, end, cfp, p);
6036 				if (error != 0)
6037 					break;
6038 				first = end;
6039 			} else {
6040 				/* handle fragment overlapped with new one */
6041 				error = nfsrv_dolocal(vp, lfp, flags,
6042 				    lop->lo_flags, first, lop->lo_end, cfp, p);
6043 				if (error != 0)
6044 					break;
6045 				first = lop->lo_end;
6046 				lop = nlop;
6047 			}
6048 		}
6049 	}
6050 	if (first < end && error == 0)
6051 		/* handle fragment past end of list */
6052 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
6053 		    end, cfp, p);
6054 
6055 	NFSEXITCODE(error);
6056 	return (error);
6057 }
6058 
6059 /*
6060  * Local lock unlock. Unlock all byte ranges that are no longer locked
6061  * by NFSv4. To do this, unlock any subranges of first-->end that
6062  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
6063  * list. This list has all locks for the file held by other
6064  * <clientid, lockowner> tuples. The list is ordered by increasing
6065  * lo_first value, but may have entries that overlap each other, for
6066  * the case of read locks.
6067  */
6068 static void
6069 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
6070     uint64_t init_end, NFSPROC_T *p)
6071 {
6072 	struct nfslock *lop;
6073 	uint64_t first, end, prevfirst __unused;
6074 
6075 	first = init_first;
6076 	end = init_end;
6077 	while (first < init_end) {
6078 		/* Loop through all nfs locks, adjusting first and end */
6079 		prevfirst = 0;
6080 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
6081 			KASSERT(prevfirst <= lop->lo_first,
6082 			    ("nfsv4 locks out of order"));
6083 			KASSERT(lop->lo_first < lop->lo_end,
6084 			    ("nfsv4 bogus lock"));
6085 			prevfirst = lop->lo_first;
6086 			if (first >= lop->lo_first &&
6087 			    first < lop->lo_end)
6088 				/*
6089 				 * Overlaps with initial part, so trim
6090 				 * off that initial part by moving first past
6091 				 * it.
6092 				 */
6093 				first = lop->lo_end;
6094 			else if (end > lop->lo_first &&
6095 			    lop->lo_first > first) {
6096 				/*
6097 				 * This lock defines the end of the
6098 				 * segment to unlock, so set end to the
6099 				 * start of it and break out of the loop.
6100 				 */
6101 				end = lop->lo_first;
6102 				break;
6103 			}
6104 			if (first >= end)
6105 				/*
6106 				 * There is no segment left to do, so
6107 				 * break out of this loop and then exit
6108 				 * the outer while() since first will be set
6109 				 * to end, which must equal init_end here.
6110 				 */
6111 				break;
6112 		}
6113 		if (first < end) {
6114 			/* Unlock this segment */
6115 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
6116 			    NFSLCK_READ, first, end, NULL, p);
6117 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
6118 			    first, end);
6119 		}
6120 		/*
6121 		 * Now move past this segment and look for any further
6122 		 * segment in the range, if there is one.
6123 		 */
6124 		first = end;
6125 		end = init_end;
6126 	}
6127 }
6128 
6129 /*
6130  * Do the local lock operation and update the rollback list, as required.
6131  * Perform the rollback and return the error if nfsvno_advlock() fails.
6132  */
6133 static int
6134 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
6135     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
6136 {
6137 	struct nfsrollback *rlp;
6138 	int error = 0, ltype, oldltype;
6139 
6140 	if (flags & NFSLCK_WRITE)
6141 		ltype = F_WRLCK;
6142 	else if (flags & NFSLCK_READ)
6143 		ltype = F_RDLCK;
6144 	else
6145 		ltype = F_UNLCK;
6146 	if (oldflags & NFSLCK_WRITE)
6147 		oldltype = F_WRLCK;
6148 	else if (oldflags & NFSLCK_READ)
6149 		oldltype = F_RDLCK;
6150 	else
6151 		oldltype = F_UNLCK;
6152 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
6153 		/* nothing to do */
6154 		goto out;
6155 	error = nfsvno_advlock(vp, ltype, first, end, p);
6156 	if (error != 0) {
6157 		if (cfp != NULL) {
6158 			cfp->cl_clientid.lval[0] = 0;
6159 			cfp->cl_clientid.lval[1] = 0;
6160 			cfp->cl_first = 0;
6161 			cfp->cl_end = NFS64BITSSET;
6162 			cfp->cl_flags = NFSLCK_WRITE;
6163 			cfp->cl_ownerlen = 5;
6164 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
6165 		}
6166 		nfsrv_locallock_rollback(vp, lfp, p);
6167 	} else if (ltype != F_UNLCK) {
6168 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
6169 		    M_WAITOK);
6170 		rlp->rlck_first = first;
6171 		rlp->rlck_end = end;
6172 		rlp->rlck_type = oldltype;
6173 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
6174 	}
6175 
6176 out:
6177 	NFSEXITCODE(error);
6178 	return (error);
6179 }
6180 
6181 /*
6182  * Roll back local lock changes and free up the rollback list.
6183  */
6184 static void
6185 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
6186 {
6187 	struct nfsrollback *rlp, *nrlp;
6188 
6189 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
6190 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
6191 		    rlp->rlck_end, p);
6192 		free(rlp, M_NFSDROLLBACK);
6193 	}
6194 	LIST_INIT(&lfp->lf_rollback);
6195 }
6196 
6197 /*
6198  * Update local lock list and delete rollback list (ie now committed to the
6199  * local locks). Most of the work is done by the internal function.
6200  */
6201 static void
6202 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
6203     uint64_t end)
6204 {
6205 	struct nfsrollback *rlp, *nrlp;
6206 	struct nfslock *new_lop, *other_lop;
6207 
6208 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
6209 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
6210 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
6211 		    M_WAITOK);
6212 	else
6213 		other_lop = NULL;
6214 	new_lop->lo_flags = flags;
6215 	new_lop->lo_first = first;
6216 	new_lop->lo_end = end;
6217 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
6218 	if (new_lop != NULL)
6219 		free(new_lop, M_NFSDLOCK);
6220 	if (other_lop != NULL)
6221 		free(other_lop, M_NFSDLOCK);
6222 
6223 	/* and get rid of the rollback list */
6224 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
6225 		free(rlp, M_NFSDROLLBACK);
6226 	LIST_INIT(&lfp->lf_rollback);
6227 }
6228 
6229 /*
6230  * Lock the struct nfslockfile for local lock updating.
6231  */
6232 static void
6233 nfsrv_locklf(struct nfslockfile *lfp)
6234 {
6235 	int gotlock;
6236 
6237 	/* lf_usecount ensures *lfp won't be free'd */
6238 	lfp->lf_usecount++;
6239 	do {
6240 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
6241 		    NFSSTATEMUTEXPTR, NULL);
6242 	} while (gotlock == 0);
6243 	lfp->lf_usecount--;
6244 }
6245 
6246 /*
6247  * Unlock the struct nfslockfile after local lock updating.
6248  */
6249 static void
6250 nfsrv_unlocklf(struct nfslockfile *lfp)
6251 {
6252 
6253 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6254 }
6255 
6256 /*
6257  * Clear out all state for the NFSv4 server.
6258  * Must be called by a thread that can sleep when no nfsds are running.
6259  */
6260 void
6261 nfsrv_throwawayallstate(NFSPROC_T *p)
6262 {
6263 	struct nfsclient *clp, *nclp;
6264 	struct nfslockfile *lfp, *nlfp;
6265 	int i;
6266 
6267 	/*
6268 	 * For each client, clean out the state and then free the structure.
6269 	 */
6270 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6271 		LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash,
6272 		    nclp) {
6273 			nfsrv_cleanclient(clp, p);
6274 			nfsrv_freedeleglist(&clp->lc_deleg);
6275 			nfsrv_freedeleglist(&clp->lc_olddeleg);
6276 			free(clp->lc_stateid, M_NFSDCLIENT);
6277 			free(clp, M_NFSDCLIENT);
6278 		}
6279 	}
6280 
6281 	/*
6282 	 * Also, free up any remaining lock file structures.
6283 	 */
6284 	for (i = 0; i < nfsrv_lockhashsize; i++) {
6285 		LIST_FOREACH_SAFE(lfp, &NFSD_VNET(nfslockhash)[i], lf_hash,
6286 		    nlfp) {
6287 			printf("nfsd unload: fnd a lock file struct\n");
6288 			nfsrv_freenfslockfile(lfp);
6289 		}
6290 	}
6291 
6292 	/* And get rid of the deviceid structures and layouts. */
6293 	nfsrv_freealllayoutsanddevids();
6294 }
6295 
6296 /*
6297  * Check the sequence# for the session and slot provided as an argument.
6298  * Also, renew the lease if the session will return NFS_OK.
6299  */
6300 int
6301 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6302     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6303     uint32_t *sflagsp, NFSPROC_T *p)
6304 {
6305 	struct nfsdsession *sep;
6306 	struct nfssessionhash *shp;
6307 	int error;
6308 
6309 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6310 	NFSLOCKSESSION(shp);
6311 	sep = nfsrv_findsession(nd->nd_sessionid);
6312 	if (sep == NULL) {
6313 		NFSUNLOCKSESSION(shp);
6314 		return (NFSERR_BADSESSION);
6315 	}
6316 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6317 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6318 	if (error != 0) {
6319 		NFSUNLOCKSESSION(shp);
6320 		return (error);
6321 	}
6322 	if (cache_this != 0)
6323 		nd->nd_flag |= ND_SAVEREPLY;
6324 	/* Renew the lease. */
6325 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6326 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6327 	nd->nd_flag |= ND_IMPLIEDCLID;
6328 
6329 	/* Handle the SP4_MECH_CRED case for NFSv4.1/4.2. */
6330 	if ((sep->sess_clp->lc_flags & LCL_MACHCRED) != 0 &&
6331 	    (nd->nd_flag & (ND_GSSINTEGRITY | ND_GSSPRIVACY)) != 0 &&
6332 	    nd->nd_princlen == sep->sess_clp->lc_namelen &&
6333 	    !NFSBCMP(sep->sess_clp->lc_name, nd->nd_principal,
6334 	    nd->nd_princlen)) {
6335 		nd->nd_flag |= ND_MACHCRED;
6336 		NFSSET_OPBIT(&nd->nd_allowops, &sep->sess_clp->lc_allowops);
6337 	}
6338 
6339 	/* Save maximum request and reply sizes. */
6340 	nd->nd_maxreq = sep->sess_maxreq;
6341 	nd->nd_maxresp = sep->sess_maxresp;
6342 
6343 	*sflagsp = 0;
6344 	if (sep->sess_clp->lc_req.nr_client == NULL ||
6345 	    (sep->sess_clp->lc_flags & LCL_CBDOWN) != 0)
6346 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
6347 	NFSUNLOCKSESSION(shp);
6348 	if (error == NFSERR_EXPIRED) {
6349 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6350 		error = 0;
6351 	} else if (error == NFSERR_ADMINREVOKED) {
6352 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6353 		error = 0;
6354 	}
6355 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6356 	return (0);
6357 }
6358 
6359 /*
6360  * Check/set reclaim complete for this session/clientid.
6361  */
6362 int
6363 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6364 {
6365 	struct nfsdsession *sep;
6366 	struct nfssessionhash *shp;
6367 	int error = 0;
6368 
6369 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6370 	NFSLOCKSTATE();
6371 	NFSLOCKSESSION(shp);
6372 	sep = nfsrv_findsession(nd->nd_sessionid);
6373 	if (sep == NULL) {
6374 		NFSUNLOCKSESSION(shp);
6375 		NFSUNLOCKSTATE();
6376 		return (NFSERR_BADSESSION);
6377 	}
6378 
6379 	if (onefs != 0)
6380 		sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6381 		/* Check to see if reclaim complete has already happened. */
6382 	else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6383 		error = NFSERR_COMPLETEALREADY;
6384 	else {
6385 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6386 		nfsrv_markreclaim(sep->sess_clp);
6387 	}
6388 	NFSUNLOCKSESSION(shp);
6389 	NFSUNLOCKSTATE();
6390 	return (error);
6391 }
6392 
6393 /*
6394  * Cache the reply in a session slot.
6395  */
6396 void
6397 nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m)
6398 {
6399 	struct nfsdsession *sep;
6400 	struct nfssessionhash *shp;
6401 	char *buf, *cp;
6402 #ifdef INET
6403 	struct sockaddr_in *sin;
6404 #endif
6405 #ifdef INET6
6406 	struct sockaddr_in6 *sin6;
6407 #endif
6408 
6409 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6410 	NFSLOCKSESSION(shp);
6411 	sep = nfsrv_findsession(nd->nd_sessionid);
6412 	if (sep == NULL) {
6413 		NFSUNLOCKSESSION(shp);
6414 		if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags &
6415 		     NFSNSF_GRACEOVER) != 0) {
6416 			buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK);
6417 			switch (nd->nd_nam->sa_family) {
6418 #ifdef INET
6419 			case AF_INET:
6420 				sin = (struct sockaddr_in *)nd->nd_nam;
6421 				cp = inet_ntop(sin->sin_family,
6422 				    &sin->sin_addr.s_addr, buf,
6423 				    INET6_ADDRSTRLEN);
6424 				break;
6425 #endif
6426 #ifdef INET6
6427 			case AF_INET6:
6428 				sin6 = (struct sockaddr_in6 *)nd->nd_nam;
6429 				cp = inet_ntop(sin6->sin6_family,
6430 				    &sin6->sin6_addr, buf, INET6_ADDRSTRLEN);
6431 				break;
6432 #endif
6433 			default:
6434 				cp = NULL;
6435 			}
6436 			if (cp != NULL)
6437 				printf("nfsrv_cache_session: no session "
6438 				    "IPaddr=%s, check NFS clients for unique "
6439 				    "/etc/hostid's\n", cp);
6440 			else
6441 				printf("nfsrv_cache_session: no session, "
6442 				    "check NFS clients for unique "
6443 				    "/etc/hostid's\n");
6444 			free(buf, M_TEMP);
6445 		}
6446 		m_freem(*m);
6447 		return;
6448 	}
6449 	nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat,
6450 	    m);
6451 	NFSUNLOCKSESSION(shp);
6452 }
6453 
6454 /*
6455  * Search for a session that matches the sessionid.
6456  */
6457 static struct nfsdsession *
6458 nfsrv_findsession(uint8_t *sessionid)
6459 {
6460 	struct nfsdsession *sep;
6461 	struct nfssessionhash *shp;
6462 
6463 	shp = NFSSESSIONHASH(sessionid);
6464 	LIST_FOREACH(sep, &shp->list, sess_hash) {
6465 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6466 			break;
6467 	}
6468 	return (sep);
6469 }
6470 
6471 /*
6472  * Destroy a session.
6473  */
6474 int
6475 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6476 {
6477 	int error, igotlock, samesess;
6478 
6479 	samesess = 0;
6480 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6481 	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6482 		samesess = 1;
6483 		if ((nd->nd_flag & ND_LASTOP) == 0)
6484 			return (NFSERR_BADSESSION);
6485 	}
6486 
6487 	/* Lock out other nfsd threads */
6488 	NFSLOCKV4ROOTMUTEX();
6489 	nfsv4_relref(&nfsv4rootfs_lock);
6490 	do {
6491 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6492 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6493 	} while (igotlock == 0);
6494 	NFSUNLOCKV4ROOTMUTEX();
6495 
6496 	error = nfsrv_freesession(nd, NULL, sessionid);
6497 	if (error == 0 && samesess != 0)
6498 		nd->nd_flag &= ~ND_HASSEQUENCE;
6499 
6500 	NFSLOCKV4ROOTMUTEX();
6501 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6502 	NFSUNLOCKV4ROOTMUTEX();
6503 	return (error);
6504 }
6505 
6506 /*
6507  * Bind a connection to a session.
6508  * For now, only certain variants are supported, since the current session
6509  * structure can only handle a single backchannel entry, which will be
6510  * applied to all connections if it is set.
6511  */
6512 int
6513 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6514 {
6515 	struct nfssessionhash *shp;
6516 	struct nfsdsession *sep;
6517 	struct nfsclient *clp;
6518 	SVCXPRT *savxprt;
6519 	int error;
6520 
6521 	error = 0;
6522 	savxprt = NULL;
6523 	shp = NFSSESSIONHASH(sessionid);
6524 	NFSLOCKSTATE();
6525 	NFSLOCKSESSION(shp);
6526 	sep = nfsrv_findsession(sessionid);
6527 	if (sep != NULL) {
6528 		clp = sep->sess_clp;
6529 		error = nfsrv_checkmachcred(NFSV4OP_BINDCONNTOSESS, nd, clp);
6530 		if (error != 0)
6531 			goto out;
6532 		if (*foreaftp == NFSCDFC4_BACK ||
6533 		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6534 		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6535 			/* Try to set up a backchannel. */
6536 			if (clp->lc_req.nr_client == NULL) {
6537 				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6538 				    "backchannel\n");
6539 				clp->lc_req.nr_client = (struct __rpc_client *)
6540 				    clnt_bck_create(nd->nd_xprt->xp_socket,
6541 				    sep->sess_cbprogram, NFSV4_CBVERS);
6542 			}
6543 			if (clp->lc_req.nr_client != NULL) {
6544 				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6545 				    "backchannel\n");
6546 				savxprt = sep->sess_cbsess.nfsess_xprt;
6547 				SVC_ACQUIRE(nd->nd_xprt);
6548 				CLNT_ACQUIRE(clp->lc_req.nr_client);
6549 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
6550 				/* Disable idle timeout. */
6551 				nd->nd_xprt->xp_idletimeout = 0;
6552 				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6553 				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6554 				clp->lc_flags |= LCL_DONEBINDCONN |
6555 				    LCL_NEEDSCBNULL;
6556 				clp->lc_flags &= ~LCL_CBDOWN;
6557 				if (*foreaftp == NFSCDFS4_BACK)
6558 					*foreaftp = NFSCDFS4_BACK;
6559 				else
6560 					*foreaftp = NFSCDFS4_BOTH;
6561 			} else if (*foreaftp != NFSCDFC4_BACK) {
6562 				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6563 				    "up backchannel\n");
6564 				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6565 				clp->lc_flags |= LCL_DONEBINDCONN;
6566 				*foreaftp = NFSCDFS4_FORE;
6567 			} else {
6568 				error = NFSERR_NOTSUPP;
6569 				printf("nfsrv_bindconnsess: Can't add "
6570 				    "backchannel\n");
6571 			}
6572 		} else {
6573 			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6574 			clp->lc_flags |= LCL_DONEBINDCONN;
6575 			*foreaftp = NFSCDFS4_FORE;
6576 		}
6577 	} else
6578 		error = NFSERR_BADSESSION;
6579 out:
6580 	NFSUNLOCKSESSION(shp);
6581 	NFSUNLOCKSTATE();
6582 	if (savxprt != NULL)
6583 		SVC_RELEASE(savxprt);
6584 	return (error);
6585 }
6586 
6587 /*
6588  * Free up a session structure.
6589  */
6590 static int
6591 nfsrv_freesession(struct nfsrv_descript *nd, struct nfsdsession *sep,
6592     uint8_t *sessionid)
6593 {
6594 	struct nfssessionhash *shp;
6595 	int i;
6596 
6597 	NFSLOCKSTATE();
6598 	if (sep == NULL) {
6599 		shp = NFSSESSIONHASH(sessionid);
6600 		NFSLOCKSESSION(shp);
6601 		sep = nfsrv_findsession(sessionid);
6602 	} else {
6603 		shp = NFSSESSIONHASH(sep->sess_sessionid);
6604 		NFSLOCKSESSION(shp);
6605 	}
6606 	if (sep != NULL) {
6607 		/* Check for the SP4_MACH_CRED case. */
6608 		if (nd != NULL && nfsrv_checkmachcred(NFSV4OP_DESTROYSESSION,
6609 		    nd, sep->sess_clp) != 0) {
6610 			NFSUNLOCKSESSION(shp);
6611 			NFSUNLOCKSTATE();
6612 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
6613 		}
6614 
6615 		sep->sess_refcnt--;
6616 		if (sep->sess_refcnt > 0) {
6617 			NFSUNLOCKSESSION(shp);
6618 			NFSUNLOCKSTATE();
6619 			return (NFSERR_BACKCHANBUSY);
6620 		}
6621 		LIST_REMOVE(sep, sess_hash);
6622 		LIST_REMOVE(sep, sess_list);
6623 	}
6624 	NFSUNLOCKSESSION(shp);
6625 	NFSUNLOCKSTATE();
6626 	if (sep == NULL)
6627 		return (NFSERR_BADSESSION);
6628 	for (i = 0; i < NFSV4_SLOTS; i++)
6629 		if (sep->sess_slots[i].nfssl_reply != NULL)
6630 			m_freem(sep->sess_slots[i].nfssl_reply);
6631 	if (sep->sess_cbsess.nfsess_xprt != NULL)
6632 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6633 	free(sep, M_NFSDSESSION);
6634 	return (0);
6635 }
6636 
6637 /*
6638  * Free a stateid.
6639  * RFC5661 says that it should fail when there are associated opens, locks
6640  * or delegations. Since stateids represent opens, I don't see how you can
6641  * free an open stateid (it will be free'd when closed), so this function
6642  * only works for lock stateids (freeing the lock_owner) or delegations.
6643  */
6644 int
6645 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6646     NFSPROC_T *p)
6647 {
6648 	struct nfsclient *clp;
6649 	struct nfsstate *stp;
6650 	int error;
6651 
6652 	NFSLOCKSTATE();
6653 	/*
6654 	 * Look up the stateid
6655 	 */
6656 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6657 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6658 	if (error == 0) {
6659 		/* First, check for a delegation. */
6660 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6661 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6662 			    NFSX_STATEIDOTHER))
6663 				break;
6664 		}
6665 		if (stp != NULL) {
6666 			nfsrv_freedeleg(stp);
6667 			NFSUNLOCKSTATE();
6668 			return (error);
6669 		}
6670 	}
6671 	/* Not a delegation, try for a lock_owner. */
6672 	if (error == 0)
6673 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6674 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6675 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6676 		/* Not a lock_owner stateid. */
6677 		error = NFSERR_LOCKSHELD;
6678 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6679 		error = NFSERR_LOCKSHELD;
6680 	if (error == 0)
6681 		nfsrv_freelockowner(stp, NULL, 0, p);
6682 	NFSUNLOCKSTATE();
6683 	return (error);
6684 }
6685 
6686 /*
6687  * Test a stateid.
6688  */
6689 int
6690 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6691     NFSPROC_T *p)
6692 {
6693 	struct nfsclient *clp;
6694 	struct nfsstate *stp;
6695 	int error;
6696 
6697 	NFSLOCKSTATE();
6698 	/*
6699 	 * Look up the stateid
6700 	 */
6701 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6702 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6703 	if (error == 0)
6704 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6705 	if (error == 0 && stateidp->seqid != 0 &&
6706 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6707 		error = NFSERR_OLDSTATEID;
6708 	NFSUNLOCKSTATE();
6709 	return (error);
6710 }
6711 
6712 /*
6713  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6714  */
6715 static int
6716 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6717     int dont_replycache, struct nfsdsession **sepp, int *slotposp)
6718 {
6719 	struct nfsdsession *sep;
6720 	uint32_t *tl, slotseq = 0;
6721 	int maxslot;
6722 	uint8_t sessionid[NFSX_V4SESSIONID];
6723 	int error;
6724 
6725 	error = nfsv4_getcbsession(clp, sepp);
6726 	if (error != 0)
6727 		return (error);
6728 	sep = *sepp;
6729 	nfsv4_sequencelookup(NULL, &sep->sess_cbsess, slotposp, &maxslot,
6730 	    &slotseq, sessionid, true);
6731 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6732 
6733 	/* Build the Sequence arguments. */
6734 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6735 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6736 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6737 	nd->nd_slotseq = tl;
6738 	nd->nd_slotid = *slotposp;
6739 	nd->nd_flag |= ND_HASSLOTID;
6740 	*tl++ = txdr_unsigned(slotseq);
6741 	*tl++ = txdr_unsigned(*slotposp);
6742 	*tl++ = txdr_unsigned(maxslot);
6743 	if (dont_replycache == 0)
6744 		*tl++ = newnfs_true;
6745 	else
6746 		*tl++ = newnfs_false;
6747 	*tl = 0;			/* No referring call list, for now. */
6748 	nd->nd_flag |= ND_HASSEQUENCE;
6749 	return (0);
6750 }
6751 
6752 /*
6753  * Get a session for the callback.
6754  */
6755 static int
6756 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6757 {
6758 	struct nfsdsession *sep;
6759 
6760 	NFSLOCKSTATE();
6761 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6762 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6763 			break;
6764 	}
6765 	if (sep == NULL) {
6766 		NFSUNLOCKSTATE();
6767 		return (NFSERR_BADSESSION);
6768 	}
6769 	sep->sess_refcnt++;
6770 	*sepp = sep;
6771 	NFSUNLOCKSTATE();
6772 	return (0);
6773 }
6774 
6775 /*
6776  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6777  * exit, since those transports will all be going away.
6778  * This is only called after all the nfsd threads are done performing RPCs,
6779  * so locking shouldn't be an issue.
6780  */
6781 void
6782 nfsrv_freeallbackchannel_xprts(void)
6783 {
6784 	struct nfsdsession *sep;
6785 	struct nfsclient *clp;
6786 	SVCXPRT *xprt;
6787 	int i;
6788 
6789 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6790 		LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
6791 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6792 				xprt = sep->sess_cbsess.nfsess_xprt;
6793 				sep->sess_cbsess.nfsess_xprt = NULL;
6794 				if (xprt != NULL)
6795 					SVC_RELEASE(xprt);
6796 			}
6797 		}
6798 	}
6799 }
6800 
6801 /*
6802  * Do a layout commit.  Actually just call nfsrv_updatemdsattr().
6803  * I have no idea if the rest of these arguments will ever be useful?
6804  */
6805 int
6806 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6807     int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6808     int hasnewmtime, struct timespec *newmtimep, int reclaim,
6809     nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6810     uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6811 {
6812 	struct nfsvattr na;
6813 	int error;
6814 
6815 	error = nfsrv_updatemdsattr(vp, &na, p);
6816 	if (error == 0) {
6817 		*hasnewsizep = 1;
6818 		*newsizep = na.na_size;
6819 	}
6820 	return (error);
6821 }
6822 
6823 /*
6824  * Try and get a layout.
6825  */
6826 int
6827 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6828     int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6829     uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6830     int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6831 {
6832 	struct nfslayouthash *lhyp;
6833 	struct nfslayout *lyp;
6834 	char *devid;
6835 	fhandle_t fh, *dsfhp;
6836 	int error, mirrorcnt;
6837 
6838 	if (nfsrv_devidcnt == 0)
6839 		return (NFSERR_UNKNLAYOUTTYPE);
6840 
6841 	if (*offset != 0)
6842 		printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6843 		    (uintmax_t)*len);
6844 	error = nfsvno_getfh(vp, &fh, p);
6845 	NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6846 	if (error != 0)
6847 		return (error);
6848 
6849 	/*
6850 	 * For now, all layouts are for entire files.
6851 	 * Only issue Read/Write layouts if requested for a non-readonly fs.
6852 	 */
6853 	if (NFSVNO_EXRDONLY(exp)) {
6854 		if (*iomode == NFSLAYOUTIOMODE_RW)
6855 			return (NFSERR_LAYOUTTRYLATER);
6856 		*iomode = NFSLAYOUTIOMODE_READ;
6857 	}
6858 	if (*iomode != NFSLAYOUTIOMODE_RW)
6859 		*iomode = NFSLAYOUTIOMODE_READ;
6860 
6861 	/*
6862 	 * Check to see if a write layout can be issued for this file.
6863 	 * This is used during mirror recovery to avoid RW layouts being
6864 	 * issued for a file while it is being copied to the recovered
6865 	 * mirror.
6866 	 */
6867 	if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6868 		return (NFSERR_LAYOUTTRYLATER);
6869 
6870 	*retonclose = 0;
6871 	*offset = 0;
6872 	*len = UINT64_MAX;
6873 
6874 	/* First, see if a layout already exists and return if found. */
6875 	lhyp = NFSLAYOUTHASH(&fh);
6876 	NFSLOCKLAYOUT(lhyp);
6877 	error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6878 	NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6879 	/*
6880 	 * Not sure if the seqid must be the same, so I won't check it.
6881 	 */
6882 	if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6883 	    stateidp->other[1] != lyp->lay_stateid.other[1] ||
6884 	    stateidp->other[2] != lyp->lay_stateid.other[2])) {
6885 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6886 			NFSUNLOCKLAYOUT(lhyp);
6887 			NFSD_DEBUG(1, "ret bad stateid\n");
6888 			return (NFSERR_BADSTATEID);
6889 		}
6890 		/*
6891 		 * I believe we get here because there is a race between
6892 		 * the client processing the CBLAYOUTRECALL and the layout
6893 		 * being deleted here on the server.
6894 		 * The client has now done a LayoutGet with a non-layout
6895 		 * stateid, as it would when there is no layout.
6896 		 * As such, free this layout and set error == NFSERR_BADSTATEID
6897 		 * so the code below will create a new layout structure as
6898 		 * would happen if no layout was found.
6899 		 * "lyp" will be set before being used below, but set it NULL
6900 		 * as a safety belt.
6901 		 */
6902 		nfsrv_freelayout(&lhyp->list, lyp);
6903 		lyp = NULL;
6904 		error = NFSERR_BADSTATEID;
6905 	}
6906 	if (error == 0) {
6907 		if (lyp->lay_layoutlen > maxcnt) {
6908 			NFSUNLOCKLAYOUT(lhyp);
6909 			NFSD_DEBUG(1, "ret layout too small\n");
6910 			return (NFSERR_TOOSMALL);
6911 		}
6912 		if (*iomode == NFSLAYOUTIOMODE_RW) {
6913 			if ((lyp->lay_flags & NFSLAY_NOSPC) != 0) {
6914 				NFSUNLOCKLAYOUT(lhyp);
6915 				NFSD_DEBUG(1, "ret layout nospace\n");
6916 				return (NFSERR_NOSPC);
6917 			}
6918 			lyp->lay_flags |= NFSLAY_RW;
6919 		} else
6920 			lyp->lay_flags |= NFSLAY_READ;
6921 		NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6922 		*layoutlenp = lyp->lay_layoutlen;
6923 		if (++lyp->lay_stateid.seqid == 0)
6924 			lyp->lay_stateid.seqid = 1;
6925 		stateidp->seqid = lyp->lay_stateid.seqid;
6926 		NFSUNLOCKLAYOUT(lhyp);
6927 		NFSD_DEBUG(4, "ret fnd layout\n");
6928 		return (0);
6929 	}
6930 	NFSUNLOCKLAYOUT(lhyp);
6931 
6932 	/* Find the device id and file handle. */
6933 	dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6934 	devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6935 	error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid);
6936 	NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6937 	if (error == 0) {
6938 		if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6939 			if (NFSX_V4FILELAYOUT > maxcnt)
6940 				error = NFSERR_TOOSMALL;
6941 			else
6942 				lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6943 				    devid, vp->v_mount->mnt_stat.f_fsid);
6944 		} else {
6945 			if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt)
6946 				error = NFSERR_TOOSMALL;
6947 			else
6948 				lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6949 				    &fh, dsfhp, devid,
6950 				    vp->v_mount->mnt_stat.f_fsid);
6951 		}
6952 	}
6953 	free(dsfhp, M_TEMP);
6954 	free(devid, M_TEMP);
6955 	if (error != 0)
6956 		return (error);
6957 
6958 	/*
6959 	 * Now, add this layout to the list.
6960 	 */
6961 	error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6962 	NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6963 	/*
6964 	 * The lyp will be set to NULL by nfsrv_addlayout() if it
6965 	 * linked the new structure into the lists.
6966 	 */
6967 	free(lyp, M_NFSDSTATE);
6968 	return (error);
6969 }
6970 
6971 /*
6972  * Generate a File Layout.
6973  */
6974 static struct nfslayout *
6975 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6976     fhandle_t *dsfhp, char *devid, fsid_t fs)
6977 {
6978 	uint32_t *tl;
6979 	struct nfslayout *lyp;
6980 	uint64_t pattern_offset;
6981 
6982 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6983 	    M_WAITOK | M_ZERO);
6984 	lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6985 	if (iomode == NFSLAYOUTIOMODE_RW)
6986 		lyp->lay_flags = NFSLAY_RW;
6987 	else
6988 		lyp->lay_flags = NFSLAY_READ;
6989 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6990 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6991 	lyp->lay_fsid = fs;
6992 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
6993 
6994 	/* Fill in the xdr for the files layout. */
6995 	tl = (uint32_t *)lyp->lay_xdr;
6996 	NFSBCOPY(devid, tl, NFSX_V4DEVICEID);		/* Device ID. */
6997 	tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6998 
6999 	/* Set the stripe size to the maximum I/O size. */
7000 	*tl++ = txdr_unsigned(nfs_srvmaxio & NFSFLAYUTIL_STRIPE_MASK);
7001 	*tl++ = 0;					/* 1st stripe index. */
7002 	pattern_offset = 0;
7003 	txdr_hyper(pattern_offset, tl); tl += 2;	/* Pattern offset. */
7004 	*tl++ = txdr_unsigned(1);			/* 1 file handle. */
7005 	*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
7006 	NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
7007 	lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
7008 	return (lyp);
7009 }
7010 
7011 #define	FLEX_OWNERID	"999"
7012 #define	FLEX_UID0	"0"
7013 /*
7014  * Generate a Flex File Layout.
7015  * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
7016  * string goes on the wire, it isn't supposed to be used by the client,
7017  * since this server uses tight coupling.
7018  * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
7019  * a string of "0". This works around the Linux Flex File Layout driver bug
7020  * which uses the synthetic uid/gid strings for the "tightly coupled" case.
7021  */
7022 static struct nfslayout *
7023 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
7024     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs)
7025 {
7026 	uint32_t *tl;
7027 	struct nfslayout *lyp;
7028 	uint64_t lenval;
7029 	int i;
7030 
7031 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt),
7032 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7033 	lyp->lay_type = NFSLAYOUT_FLEXFILE;
7034 	if (iomode == NFSLAYOUTIOMODE_RW)
7035 		lyp->lay_flags = NFSLAY_RW;
7036 	else
7037 		lyp->lay_flags = NFSLAY_READ;
7038 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
7039 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
7040 	lyp->lay_fsid = fs;
7041 	lyp->lay_mirrorcnt = mirrorcnt;
7042 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
7043 
7044 	/* Fill in the xdr for the files layout. */
7045 	tl = (uint32_t *)lyp->lay_xdr;
7046 	lenval = 0;
7047 	txdr_hyper(lenval, tl); tl += 2;		/* Stripe unit. */
7048 	*tl++ = txdr_unsigned(mirrorcnt);		/* # of mirrors. */
7049 	for (i = 0; i < mirrorcnt; i++) {
7050 		*tl++ = txdr_unsigned(1);		/* One stripe. */
7051 		NFSBCOPY(devid, tl, NFSX_V4DEVICEID);	/* Device ID. */
7052 		tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7053 		devid += NFSX_V4DEVICEID;
7054 		*tl++ = txdr_unsigned(1);		/* Efficiency. */
7055 		*tl++ = 0;				/* Proxy Stateid. */
7056 		*tl++ = 0x55555555;
7057 		*tl++ = 0x55555555;
7058 		*tl++ = 0x55555555;
7059 		*tl++ = txdr_unsigned(1);		/* 1 file handle. */
7060 		*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
7061 		NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
7062 		tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
7063 		dsfhp++;
7064 		if (nfsrv_flexlinuxhack != 0) {
7065 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
7066 			*tl = 0;		/* 0 pad string. */
7067 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
7068 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
7069 			*tl = 0;		/* 0 pad string. */
7070 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
7071 		} else {
7072 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
7073 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
7074 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
7075 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
7076 		}
7077 	}
7078 	*tl++ = txdr_unsigned(0);		/* ff_flags. */
7079 	*tl = txdr_unsigned(60);		/* Status interval hint. */
7080 	lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt);
7081 	return (lyp);
7082 }
7083 
7084 /*
7085  * Parse and process Flex File errors returned via LayoutReturn.
7086  */
7087 static void
7088 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
7089     NFSPROC_T *p)
7090 {
7091 	uint32_t *tl;
7092 	int cnt, errcnt, i, j, opnum, stat;
7093 	char devid[NFSX_V4DEVICEID];
7094 
7095 	tl = layp;
7096 	maxcnt -= NFSX_UNSIGNED;
7097 	if (maxcnt > 0)
7098 		cnt = fxdr_unsigned(int, *tl++);
7099 	else
7100 		cnt = 0;
7101 	NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
7102 	for (i = 0; i < cnt; i++) {
7103 		maxcnt -= NFSX_STATEID + 2 * NFSX_HYPER +
7104 		    NFSX_UNSIGNED;
7105 		if (maxcnt <= 0)
7106 			break;
7107 		/* Skip offset, length and stateid for now. */
7108 		tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
7109 		errcnt = fxdr_unsigned(int, *tl++);
7110 		NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
7111 		for (j = 0; j < errcnt; j++) {
7112 			maxcnt -= NFSX_V4DEVICEID + 2 * NFSX_UNSIGNED;
7113 			if (maxcnt < 0)
7114 				break;
7115 			NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
7116 			tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7117 			stat = fxdr_unsigned(int, *tl++);
7118 			opnum = fxdr_unsigned(int, *tl++);
7119 			NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
7120 			    stat);
7121 			/*
7122 			 * Except for NFSERR_ACCES, NFSERR_STALE and
7123 			 * NFSERR_NOSPC errors, disable the mirror.
7124 			 */
7125 			if (stat != NFSERR_ACCES && stat != NFSERR_STALE &&
7126 			    stat != NFSERR_NOSPC)
7127 				nfsrv_delds(devid, p);
7128 
7129 			/* For NFSERR_NOSPC, mark all devids and layouts. */
7130 			if (stat == NFSERR_NOSPC)
7131 				nfsrv_marknospc(devid, true);
7132 		}
7133 	}
7134 }
7135 
7136 /*
7137  * This function removes all flex file layouts which has a mirror with
7138  * a device id that matches the argument.
7139  * Called when the DS represented by the device id has failed.
7140  */
7141 void
7142 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
7143 {
7144 	uint32_t *tl;
7145 	struct nfslayout *lyp, *nlyp;
7146 	struct nfslayouthash *lhyp;
7147 	struct nfslayouthead loclyp;
7148 	int i, j;
7149 
7150 	NFSD_DEBUG(4, "flexmirrordel\n");
7151 	/* Move all layouts found onto a local list. */
7152 	TAILQ_INIT(&loclyp);
7153 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7154 		lhyp = &nfslayouthash[i];
7155 		NFSLOCKLAYOUT(lhyp);
7156 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7157 			if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
7158 			    lyp->lay_mirrorcnt > 1) {
7159 				NFSD_DEBUG(4, "possible match\n");
7160 				tl = lyp->lay_xdr;
7161 				tl += 3;
7162 				for (j = 0; j < lyp->lay_mirrorcnt; j++) {
7163 					tl++;
7164 					if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
7165 					    == 0) {
7166 						/* Found one. */
7167 						NFSD_DEBUG(4, "fnd one\n");
7168 						TAILQ_REMOVE(&lhyp->list, lyp,
7169 						    lay_list);
7170 						TAILQ_INSERT_HEAD(&loclyp, lyp,
7171 						    lay_list);
7172 						break;
7173 					}
7174 					tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
7175 					    NFSM_RNDUP(NFSX_V4PNFSFH) /
7176 					    NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
7177 				}
7178 			}
7179 		}
7180 		NFSUNLOCKLAYOUT(lhyp);
7181 	}
7182 
7183 	/* Now, try to do a Layout recall for each one found. */
7184 	TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
7185 		NFSD_DEBUG(4, "do layout recall\n");
7186 		/*
7187 		 * The layout stateid.seqid needs to be incremented
7188 		 * before doing a LAYOUT_RECALL callback.
7189 		 */
7190 		if (++lyp->lay_stateid.seqid == 0)
7191 			lyp->lay_stateid.seqid = 1;
7192 		nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
7193 		    &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
7194 		nfsrv_freelayout(&loclyp, lyp);
7195 	}
7196 }
7197 
7198 /*
7199  * Do a recall callback to the client for this layout.
7200  */
7201 static int
7202 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
7203     struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
7204 {
7205 	struct nfsclient *clp;
7206 	int error;
7207 
7208 	NFSD_DEBUG(4, "nfsrv_recalllayout\n");
7209 	error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
7210 	    0, NULL, p);
7211 	NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
7212 	if (error != 0) {
7213 		printf("nfsrv_recalllayout: getclient err=%d\n", error);
7214 		return (error);
7215 	}
7216 	if ((clp->lc_flags & LCL_NFSV41) != 0) {
7217 		error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
7218 		    stateidp, changed, fhp, NULL, NULL, laytype, p);
7219 		/* If lyp != NULL, handle an error return here. */
7220 		if (error != 0 && lyp != NULL) {
7221 			NFSDRECALLLOCK();
7222 			/*
7223 			 * Mark it returned, since no layout recall
7224 			 * has been done.
7225 			 * All errors seem to be non-recoverable, although
7226 			 * NFSERR_NOMATCHLAYOUT is a normal event.
7227 			 */
7228 			if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
7229 				lyp->lay_flags |= NFSLAY_RETURNED;
7230 				wakeup(lyp);
7231 			}
7232 			NFSDRECALLUNLOCK();
7233 			if (error != NFSERR_NOMATCHLAYOUT)
7234 				printf("nfsrv_recalllayout: err=%d\n", error);
7235 		}
7236 	} else
7237 		printf("nfsrv_recalllayout: clp not NFSv4.1\n");
7238 	return (error);
7239 }
7240 
7241 /*
7242  * Find a layout to recall when we exceed our high water mark.
7243  */
7244 void
7245 nfsrv_recalloldlayout(NFSPROC_T *p)
7246 {
7247 	struct nfslayouthash *lhyp;
7248 	struct nfslayout *lyp;
7249 	nfsquad_t clientid;
7250 	nfsv4stateid_t stateid;
7251 	fhandle_t fh;
7252 	int error, laytype = 0, ret;
7253 
7254 	lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
7255 	NFSLOCKLAYOUT(lhyp);
7256 	TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
7257 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
7258 			lyp->lay_flags |= NFSLAY_CALLB;
7259 			/*
7260 			 * The layout stateid.seqid needs to be incremented
7261 			 * before doing a LAYOUT_RECALL callback.
7262 			 */
7263 			if (++lyp->lay_stateid.seqid == 0)
7264 				lyp->lay_stateid.seqid = 1;
7265 			clientid = lyp->lay_clientid;
7266 			stateid = lyp->lay_stateid;
7267 			NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
7268 			laytype = lyp->lay_type;
7269 			break;
7270 		}
7271 	}
7272 	NFSUNLOCKLAYOUT(lhyp);
7273 	if (lyp != NULL) {
7274 		error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
7275 		    laytype, p);
7276 		if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
7277 			NFSD_DEBUG(4, "recallold=%d\n", error);
7278 		if (error != 0) {
7279 			NFSLOCKLAYOUT(lhyp);
7280 			/*
7281 			 * Since the hash list was unlocked, we need to
7282 			 * find it again.
7283 			 */
7284 			ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
7285 			    &lyp);
7286 			if (ret == 0 &&
7287 			    (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
7288 			    lyp->lay_stateid.other[0] == stateid.other[0] &&
7289 			    lyp->lay_stateid.other[1] == stateid.other[1] &&
7290 			    lyp->lay_stateid.other[2] == stateid.other[2]) {
7291 				/*
7292 				 * The client no longer knows this layout, so
7293 				 * it can be free'd now.
7294 				 */
7295 				if (error == NFSERR_NOMATCHLAYOUT)
7296 					nfsrv_freelayout(&lhyp->list, lyp);
7297 				else {
7298 					/*
7299 					 * Leave it to be tried later by
7300 					 * clearing NFSLAY_CALLB and moving
7301 					 * it to the head of the list, so it
7302 					 * won't be tried again for a while.
7303 					 */
7304 					lyp->lay_flags &= ~NFSLAY_CALLB;
7305 					TAILQ_REMOVE(&lhyp->list, lyp,
7306 					    lay_list);
7307 					TAILQ_INSERT_HEAD(&lhyp->list, lyp,
7308 					    lay_list);
7309 				}
7310 			}
7311 			NFSUNLOCKLAYOUT(lhyp);
7312 		}
7313 	}
7314 }
7315 
7316 /*
7317  * Try and return layout(s).
7318  */
7319 int
7320 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
7321     int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
7322     int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
7323     struct ucred *cred, NFSPROC_T *p)
7324 {
7325 	struct nfsvattr na;
7326 	struct nfslayouthash *lhyp;
7327 	struct nfslayout *lyp;
7328 	fhandle_t fh;
7329 	int error = 0;
7330 
7331 	*fndp = 0;
7332 	if (kind == NFSV4LAYOUTRET_FILE) {
7333 		error = nfsvno_getfh(vp, &fh, p);
7334 		if (error == 0) {
7335 			error = nfsrv_updatemdsattr(vp, &na, p);
7336 			if (error != 0)
7337 				printf("nfsrv_layoutreturn: updatemdsattr"
7338 				    " failed=%d\n", error);
7339 		}
7340 		if (error == 0) {
7341 			if (reclaim == newnfs_true) {
7342 				error = nfsrv_checkgrace(NULL, NULL,
7343 				    NFSLCK_RECLAIM);
7344 				if (error != NFSERR_NOGRACE)
7345 					error = 0;
7346 				return (error);
7347 			}
7348 			lhyp = NFSLAYOUTHASH(&fh);
7349 			NFSDRECALLLOCK();
7350 			NFSLOCKLAYOUT(lhyp);
7351 			error = nfsrv_findlayout(&nd->nd_clientid, &fh,
7352 			    layouttype, p, &lyp);
7353 			NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
7354 			if (error == 0 &&
7355 			    stateidp->other[0] == lyp->lay_stateid.other[0] &&
7356 			    stateidp->other[1] == lyp->lay_stateid.other[1] &&
7357 			    stateidp->other[2] == lyp->lay_stateid.other[2]) {
7358 				NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
7359 				    " %x %x %x laystateid %d %x %x %x"
7360 				    " off=%ju len=%ju flgs=0x%x\n",
7361 				    stateidp->seqid, stateidp->other[0],
7362 				    stateidp->other[1], stateidp->other[2],
7363 				    lyp->lay_stateid.seqid,
7364 				    lyp->lay_stateid.other[0],
7365 				    lyp->lay_stateid.other[1],
7366 				    lyp->lay_stateid.other[2],
7367 				    (uintmax_t)offset, (uintmax_t)len,
7368 				    lyp->lay_flags);
7369 				if (++lyp->lay_stateid.seqid == 0)
7370 					lyp->lay_stateid.seqid = 1;
7371 				stateidp->seqid = lyp->lay_stateid.seqid;
7372 				if (offset == 0 && len == UINT64_MAX) {
7373 					if ((iomode & NFSLAYOUTIOMODE_READ) !=
7374 					    0)
7375 						lyp->lay_flags &= ~NFSLAY_READ;
7376 					if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7377 						lyp->lay_flags &= ~NFSLAY_RW;
7378 					if ((lyp->lay_flags & (NFSLAY_READ |
7379 					    NFSLAY_RW)) == 0)
7380 						nfsrv_freelayout(&lhyp->list,
7381 						    lyp);
7382 					else
7383 						*fndp = 1;
7384 				} else
7385 					*fndp = 1;
7386 			}
7387 			NFSUNLOCKLAYOUT(lhyp);
7388 			/* Search the nfsrv_recalllist for a match. */
7389 			TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7390 				if (NFSBCMP(&lyp->lay_fh, &fh,
7391 				    sizeof(fh)) == 0 &&
7392 				    lyp->lay_clientid.qval ==
7393 				    nd->nd_clientid.qval &&
7394 				    stateidp->other[0] ==
7395 				    lyp->lay_stateid.other[0] &&
7396 				    stateidp->other[1] ==
7397 				    lyp->lay_stateid.other[1] &&
7398 				    stateidp->other[2] ==
7399 				    lyp->lay_stateid.other[2]) {
7400 					lyp->lay_flags |= NFSLAY_RETURNED;
7401 					wakeup(lyp);
7402 					error = 0;
7403 				}
7404 			}
7405 			NFSDRECALLUNLOCK();
7406 		}
7407 		if (layouttype == NFSLAYOUT_FLEXFILE && layp != NULL)
7408 			nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7409 	} else if (kind == NFSV4LAYOUTRET_FSID)
7410 		nfsrv_freelayouts(&nd->nd_clientid,
7411 		    &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7412 	else if (kind == NFSV4LAYOUTRET_ALL)
7413 		nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7414 	else
7415 		error = NFSERR_INVAL;
7416 	if (error == -1)
7417 		error = 0;
7418 	return (error);
7419 }
7420 
7421 /*
7422  * Look for an existing layout.
7423  */
7424 static int
7425 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7426     NFSPROC_T *p, struct nfslayout **lypp)
7427 {
7428 	struct nfslayouthash *lhyp;
7429 	struct nfslayout *lyp;
7430 	int ret;
7431 
7432 	*lypp = NULL;
7433 	ret = 0;
7434 	lhyp = NFSLAYOUTHASH(fhp);
7435 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7436 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7437 		    lyp->lay_clientid.qval == clientidp->qval &&
7438 		    lyp->lay_type == laytype)
7439 			break;
7440 	}
7441 	if (lyp != NULL)
7442 		*lypp = lyp;
7443 	else
7444 		ret = -1;
7445 	return (ret);
7446 }
7447 
7448 /*
7449  * Add the new layout, as required.
7450  */
7451 static int
7452 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7453     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7454 {
7455 	struct nfsclient *clp;
7456 	struct nfslayouthash *lhyp;
7457 	struct nfslayout *lyp, *nlyp;
7458 	fhandle_t *fhp;
7459 	int error;
7460 
7461 	KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7462 	    ("nfsrv_layoutget: no nd_clientid\n"));
7463 	lyp = *lypp;
7464 	fhp = &lyp->lay_fh;
7465 	NFSLOCKSTATE();
7466 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7467 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7468 	if (error != 0) {
7469 		NFSUNLOCKSTATE();
7470 		return (error);
7471 	}
7472 	lyp->lay_stateid.seqid = stateidp->seqid = 1;
7473 	lyp->lay_stateid.other[0] = stateidp->other[0] =
7474 	    clp->lc_clientid.lval[0];
7475 	lyp->lay_stateid.other[1] = stateidp->other[1] =
7476 	    clp->lc_clientid.lval[1];
7477 	lyp->lay_stateid.other[2] = stateidp->other[2] =
7478 	    nfsrv_nextstateindex(clp);
7479 	NFSUNLOCKSTATE();
7480 
7481 	lhyp = NFSLAYOUTHASH(fhp);
7482 	NFSLOCKLAYOUT(lhyp);
7483 	TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7484 		if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7485 		    nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7486 			break;
7487 	}
7488 	if (nlyp != NULL) {
7489 		/* A layout already exists, so use it. */
7490 		nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7491 		NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7492 		*layoutlenp = nlyp->lay_layoutlen;
7493 		if (++nlyp->lay_stateid.seqid == 0)
7494 			nlyp->lay_stateid.seqid = 1;
7495 		stateidp->seqid = nlyp->lay_stateid.seqid;
7496 		stateidp->other[0] = nlyp->lay_stateid.other[0];
7497 		stateidp->other[1] = nlyp->lay_stateid.other[1];
7498 		stateidp->other[2] = nlyp->lay_stateid.other[2];
7499 		NFSUNLOCKLAYOUT(lhyp);
7500 		return (0);
7501 	}
7502 
7503 	/* Insert the new layout in the lists. */
7504 	*lypp = NULL;
7505 	atomic_add_int(&nfsrv_layoutcnt, 1);
7506 	NFSD_VNET(nfsstatsv1_p)->srvlayouts++;
7507 	NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7508 	*layoutlenp = lyp->lay_layoutlen;
7509 	TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7510 	NFSUNLOCKLAYOUT(lhyp);
7511 	return (0);
7512 }
7513 
7514 /*
7515  * Get the devinfo for a deviceid.
7516  */
7517 int
7518 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7519     uint32_t *notify, int *devaddrlen, char **devaddr)
7520 {
7521 	struct nfsdevice *ds;
7522 
7523 	if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7524 	     NFSLAYOUT_FLEXFILE) ||
7525 	    (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7526 		return (NFSERR_UNKNLAYOUTTYPE);
7527 
7528 	/*
7529 	 * Now, search for the device id.  Note that the structures won't go
7530 	 * away, but the order changes in the list.  As such, the lock only
7531 	 * needs to be held during the search through the list.
7532 	 */
7533 	NFSDDSLOCK();
7534 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7535 		if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7536 		    ds->nfsdev_nmp != NULL)
7537 			break;
7538 	}
7539 	NFSDDSUNLOCK();
7540 	if (ds == NULL)
7541 		return (NFSERR_NOENT);
7542 
7543 	/* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7544 	*devaddrlen = 0;
7545 	if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7546 		*devaddrlen = ds->nfsdev_fileaddrlen;
7547 		*devaddr = ds->nfsdev_fileaddr;
7548 	} else if (layouttype == NFSLAYOUT_FLEXFILE) {
7549 		*devaddrlen = ds->nfsdev_flexaddrlen;
7550 		*devaddr = ds->nfsdev_flexaddr;
7551 	}
7552 	if (*devaddrlen == 0)
7553 		return (NFSERR_UNKNLAYOUTTYPE);
7554 
7555 	/*
7556 	 * The XDR overhead is 3 unsigned values: layout_type,
7557 	 * length_of_address and notify bitmap.
7558 	 * If the notify array is changed to not all zeros, the
7559 	 * count of unsigned values must be increased.
7560 	 */
7561 	if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7562 	    3 * NFSX_UNSIGNED) {
7563 		*maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7564 		return (NFSERR_TOOSMALL);
7565 	}
7566 	return (0);
7567 }
7568 
7569 /*
7570  * Free a list of layout state structures.
7571  */
7572 static void
7573 nfsrv_freelayoutlist(nfsquad_t clientid)
7574 {
7575 	struct nfslayouthash *lhyp;
7576 	struct nfslayout *lyp, *nlyp;
7577 	int i;
7578 
7579 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7580 		lhyp = &nfslayouthash[i];
7581 		NFSLOCKLAYOUT(lhyp);
7582 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7583 			if (lyp->lay_clientid.qval == clientid.qval)
7584 				nfsrv_freelayout(&lhyp->list, lyp);
7585 		}
7586 		NFSUNLOCKLAYOUT(lhyp);
7587 	}
7588 }
7589 
7590 /*
7591  * Free up a layout.
7592  */
7593 static void
7594 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7595 {
7596 
7597 	NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7598 	atomic_add_int(&nfsrv_layoutcnt, -1);
7599 	NFSD_VNET(nfsstatsv1_p)->srvlayouts--;
7600 	TAILQ_REMOVE(lhp, lyp, lay_list);
7601 	free(lyp, M_NFSDSTATE);
7602 }
7603 
7604 /*
7605  * Free up a device id.
7606  */
7607 void
7608 nfsrv_freeonedevid(struct nfsdevice *ds)
7609 {
7610 	int i;
7611 
7612 	atomic_add_int(&nfsrv_devidcnt, -1);
7613 	vrele(ds->nfsdev_dvp);
7614 	for (i = 0; i < nfsrv_dsdirsize; i++)
7615 		if (ds->nfsdev_dsdir[i] != NULL)
7616 			vrele(ds->nfsdev_dsdir[i]);
7617 	free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7618 	free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7619 	free(ds->nfsdev_host, M_NFSDSTATE);
7620 	free(ds, M_NFSDSTATE);
7621 }
7622 
7623 /*
7624  * Free up a device id and its mirrors.
7625  */
7626 static void
7627 nfsrv_freedevid(struct nfsdevice *ds)
7628 {
7629 
7630 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7631 	nfsrv_freeonedevid(ds);
7632 }
7633 
7634 /*
7635  * Free all layouts and device ids.
7636  * Done when the nfsd threads are shut down since there may be a new
7637  * modified device id list created when the nfsd is restarted.
7638  */
7639 void
7640 nfsrv_freealllayoutsanddevids(void)
7641 {
7642 	struct nfsdontlist *mrp, *nmrp;
7643 	struct nfslayout *lyp, *nlyp;
7644 
7645 	/* Get rid of the deviceid structures. */
7646 	nfsrv_freealldevids();
7647 	TAILQ_INIT(&nfsrv_devidhead);
7648 	nfsrv_devidcnt = 0;
7649 
7650 	/* Get rid of all layouts. */
7651 	nfsrv_freealllayouts();
7652 
7653 	/* Get rid of any nfsdontlist entries. */
7654 	LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7655 		free(mrp, M_NFSDSTATE);
7656 	LIST_INIT(&nfsrv_dontlisthead);
7657 	nfsrv_dontlistlen = 0;
7658 
7659 	/* Free layouts in the recall list. */
7660 	TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7661 		nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7662 	TAILQ_INIT(&nfsrv_recalllisthead);
7663 }
7664 
7665 /*
7666  * Free layouts that match the arguments.
7667  */
7668 static void
7669 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7670 {
7671 	struct nfslayouthash *lhyp;
7672 	struct nfslayout *lyp, *nlyp;
7673 	int i;
7674 
7675 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7676 		lhyp = &nfslayouthash[i];
7677 		NFSLOCKLAYOUT(lhyp);
7678 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7679 			if (clid->qval != lyp->lay_clientid.qval)
7680 				continue;
7681 			if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0)
7682 				continue;
7683 			if (laytype != lyp->lay_type)
7684 				continue;
7685 			if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7686 				lyp->lay_flags &= ~NFSLAY_READ;
7687 			if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7688 				lyp->lay_flags &= ~NFSLAY_RW;
7689 			if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7690 				nfsrv_freelayout(&lhyp->list, lyp);
7691 		}
7692 		NFSUNLOCKLAYOUT(lhyp);
7693 	}
7694 }
7695 
7696 /*
7697  * Free all layouts for the argument file.
7698  */
7699 void
7700 nfsrv_freefilelayouts(fhandle_t *fhp)
7701 {
7702 	struct nfslayouthash *lhyp;
7703 	struct nfslayout *lyp, *nlyp;
7704 
7705 	lhyp = NFSLAYOUTHASH(fhp);
7706 	NFSLOCKLAYOUT(lhyp);
7707 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7708 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7709 			nfsrv_freelayout(&lhyp->list, lyp);
7710 	}
7711 	NFSUNLOCKLAYOUT(lhyp);
7712 }
7713 
7714 /*
7715  * Free all layouts.
7716  */
7717 static void
7718 nfsrv_freealllayouts(void)
7719 {
7720 	struct nfslayouthash *lhyp;
7721 	struct nfslayout *lyp, *nlyp;
7722 	int i;
7723 
7724 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7725 		lhyp = &nfslayouthash[i];
7726 		NFSLOCKLAYOUT(lhyp);
7727 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7728 			nfsrv_freelayout(&lhyp->list, lyp);
7729 		NFSUNLOCKLAYOUT(lhyp);
7730 	}
7731 }
7732 
7733 /*
7734  * Look up the mount path for the DS server.
7735  */
7736 static int
7737 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7738     struct nfsdevice **dsp)
7739 {
7740 	struct nameidata nd;
7741 	struct nfsdevice *ds;
7742 	struct mount *mp;
7743 	int error, i;
7744 	char *dsdirpath;
7745 	size_t dsdirsize;
7746 
7747 	NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7748 	*dsp = NULL;
7749 	if (jailed(p->td_ucred)) {
7750 		printf("A pNFS nfsd cannot run in a jail\n");
7751 		return (EPERM);
7752 	}
7753 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7754 	    dspathp);
7755 	error = namei(&nd);
7756 	NFSD_DEBUG(4, "lookup=%d\n", error);
7757 	if (error != 0)
7758 		return (error);
7759 	if (nd.ni_vp->v_type != VDIR) {
7760 		vput(nd.ni_vp);
7761 		NFSD_DEBUG(4, "dspath not dir\n");
7762 		return (ENOTDIR);
7763 	}
7764 	if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7765 		vput(nd.ni_vp);
7766 		NFSD_DEBUG(4, "dspath not an NFS mount\n");
7767 		return (ENXIO);
7768 	}
7769 
7770 	/*
7771 	 * Allocate a DS server structure with the NFS mounted directory
7772 	 * vnode reference counted, so that a non-forced dismount will
7773 	 * fail with EBUSY.
7774 	 * This structure is always linked into the list, even if an error
7775 	 * is being returned.  The caller will free the entire list upon
7776 	 * an error return.
7777 	 */
7778 	*dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7779 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7780 	ds->nfsdev_dvp = nd.ni_vp;
7781 	ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7782 	NFSVOPUNLOCK(nd.ni_vp);
7783 
7784 	dsdirsize = strlen(dspathp) + 16;
7785 	dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7786 	/* Now, create the DS directory structures. */
7787 	for (i = 0; i < nfsrv_dsdirsize; i++) {
7788 		snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7789 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7790 		    UIO_SYSSPACE, dsdirpath);
7791 		error = namei(&nd);
7792 		NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7793 		if (error != 0)
7794 			break;
7795 		if (nd.ni_vp->v_type != VDIR) {
7796 			vput(nd.ni_vp);
7797 			error = ENOTDIR;
7798 			NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7799 			break;
7800 		}
7801 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7802 			vput(nd.ni_vp);
7803 			error = ENXIO;
7804 			NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7805 			break;
7806 		}
7807 		ds->nfsdev_dsdir[i] = nd.ni_vp;
7808 		NFSVOPUNLOCK(nd.ni_vp);
7809 	}
7810 	free(dsdirpath, M_TEMP);
7811 
7812 	if (strlen(mdspathp) > 0) {
7813 		/*
7814 		 * This DS stores file for a specific MDS exported file
7815 		 * system.
7816 		 */
7817 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7818 		    UIO_SYSSPACE, mdspathp);
7819 		error = namei(&nd);
7820 		NFSD_DEBUG(4, "mds lookup=%d\n", error);
7821 		if (error != 0)
7822 			goto out;
7823 		if (nd.ni_vp->v_type != VDIR) {
7824 			vput(nd.ni_vp);
7825 			error = ENOTDIR;
7826 			NFSD_DEBUG(4, "mdspath not dir\n");
7827 			goto out;
7828 		}
7829 		mp = nd.ni_vp->v_mount;
7830 		if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7831 			vput(nd.ni_vp);
7832 			error = ENXIO;
7833 			NFSD_DEBUG(4, "mdspath not an exported fs\n");
7834 			goto out;
7835 		}
7836 		ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7837 		ds->nfsdev_mdsisset = 1;
7838 		vput(nd.ni_vp);
7839 	}
7840 
7841 out:
7842 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7843 	atomic_add_int(&nfsrv_devidcnt, 1);
7844 	return (error);
7845 }
7846 
7847 /*
7848  * Look up the mount path for the DS server and delete it.
7849  */
7850 int
7851 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7852 {
7853 	struct mount *mp;
7854 	struct nfsmount *nmp;
7855 	struct nfsdevice *ds;
7856 	int error;
7857 
7858 	NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7859 	/*
7860 	 * Search for the path in the mount list.  Avoid looking the path
7861 	 * up, since this mount point may be hung, with associated locked
7862 	 * vnodes, etc.
7863 	 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7864 	 * until this completes.
7865 	 * As noted in the man page, this should be done before any forced
7866 	 * dismount on the mount point, but at least the handshake on
7867 	 * NFSMNTP_CANCELRPCS should make it safe.
7868 	 */
7869 	error = 0;
7870 	ds = NULL;
7871 	nmp = NULL;
7872 	mtx_lock(&mountlist_mtx);
7873 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7874 		if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7875 		    strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7876 		    mp->mnt_data != NULL) {
7877 			nmp = VFSTONFS(mp);
7878 			NFSLOCKMNT(nmp);
7879 			if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7880 			     NFSMNTP_CANCELRPCS)) == 0) {
7881 				nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7882 				NFSUNLOCKMNT(nmp);
7883 			} else {
7884 				NFSUNLOCKMNT(nmp);
7885 				nmp = NULL;
7886 			}
7887 			break;
7888 		}
7889 	}
7890 	mtx_unlock(&mountlist_mtx);
7891 
7892 	if (nmp != NULL) {
7893 		ds = nfsrv_deldsnmp(op, nmp, p);
7894 		NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7895 		if (ds != NULL) {
7896 			nfsrv_killrpcs(nmp);
7897 			NFSD_DEBUG(4, "aft killrpcs\n");
7898 		} else
7899 			error = ENXIO;
7900 		NFSLOCKMNT(nmp);
7901 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7902 		wakeup(nmp);
7903 		NFSUNLOCKMNT(nmp);
7904 	} else
7905 		error = EINVAL;
7906 	return (error);
7907 }
7908 
7909 /*
7910  * Search for and remove a DS entry which matches the "nmp" argument.
7911  * The nfsdevice structure pointer is returned so that the caller can
7912  * free it via nfsrv_freeonedevid().
7913  * For the forced case, do not try to do LayoutRecalls, since the server
7914  * must be shut down now anyhow.
7915  */
7916 struct nfsdevice *
7917 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7918 {
7919 	struct nfsdevice *fndds;
7920 
7921 	NFSD_DEBUG(4, "deldsdvp\n");
7922 	NFSDDSLOCK();
7923 	if (op == PNFSDOP_FORCEDELDS)
7924 		fndds = nfsv4_findmirror(nmp);
7925 	else
7926 		fndds = nfsrv_findmirroredds(nmp);
7927 	if (fndds != NULL)
7928 		nfsrv_deleteds(fndds);
7929 	NFSDDSUNLOCK();
7930 	if (fndds != NULL) {
7931 		if (op != PNFSDOP_FORCEDELDS)
7932 			nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7933 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7934 	}
7935 	return (fndds);
7936 }
7937 
7938 /*
7939  * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7940  * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7941  * point.
7942  * Also, returns an error instead of the nfsdevice found.
7943  */
7944 int
7945 nfsrv_delds(char *devid, NFSPROC_T *p)
7946 {
7947 	struct nfsdevice *ds, *fndds;
7948 	struct nfsmount *nmp;
7949 	int fndmirror;
7950 
7951 	NFSD_DEBUG(4, "delds\n");
7952 	/*
7953 	 * Search the DS server list for a match with devid.
7954 	 * Remove the DS entry if found and there is a mirror.
7955 	 */
7956 	fndds = NULL;
7957 	nmp = NULL;
7958 	fndmirror = 0;
7959 	NFSDDSLOCK();
7960 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7961 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7962 		    ds->nfsdev_nmp != NULL) {
7963 			NFSD_DEBUG(4, "fnd main ds\n");
7964 			fndds = ds;
7965 			break;
7966 		}
7967 	}
7968 	if (fndds == NULL) {
7969 		NFSDDSUNLOCK();
7970 		return (ENXIO);
7971 	}
7972 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7973 		fndmirror = 1;
7974 	else if (fndds->nfsdev_mdsisset != 0) {
7975 		/* For the fsid is set case, search for a mirror. */
7976 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7977 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
7978 			    ds->nfsdev_mdsisset != 0 &&
7979 			    fsidcmp(&ds->nfsdev_mdsfsid,
7980 			    &fndds->nfsdev_mdsfsid) == 0) {
7981 				fndmirror = 1;
7982 				break;
7983 			}
7984 		}
7985 	}
7986 	if (fndmirror != 0) {
7987 		nmp = fndds->nfsdev_nmp;
7988 		NFSLOCKMNT(nmp);
7989 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7990 		     NFSMNTP_CANCELRPCS)) == 0) {
7991 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7992 			NFSUNLOCKMNT(nmp);
7993 			nfsrv_deleteds(fndds);
7994 		} else {
7995 			NFSUNLOCKMNT(nmp);
7996 			nmp = NULL;
7997 		}
7998 	}
7999 	NFSDDSUNLOCK();
8000 	if (nmp != NULL) {
8001 		nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
8002 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
8003 		nfsrv_killrpcs(nmp);
8004 		NFSLOCKMNT(nmp);
8005 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
8006 		wakeup(nmp);
8007 		NFSUNLOCKMNT(nmp);
8008 		return (0);
8009 	}
8010 	return (ENXIO);
8011 }
8012 
8013 /*
8014  * Mark a DS as disabled by setting nfsdev_nmp = NULL.
8015  */
8016 static void
8017 nfsrv_deleteds(struct nfsdevice *fndds)
8018 {
8019 
8020 	NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
8021 	fndds->nfsdev_nmp = NULL;
8022 	if (fndds->nfsdev_mdsisset == 0)
8023 		nfsrv_faildscnt--;
8024 }
8025 
8026 /*
8027  * Fill in the addr structures for the File and Flex File layouts.
8028  */
8029 static void
8030 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
8031 {
8032 	uint32_t *tl;
8033 	char *netprot;
8034 	int addrlen;
8035 	static uint64_t new_devid = 0;
8036 
8037 	if (strchr(addr, ':') != NULL)
8038 		netprot = "tcp6";
8039 	else
8040 		netprot = "tcp";
8041 
8042 	/* Fill in the device id. */
8043 	NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
8044 	new_devid++;
8045 	NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
8046 	    sizeof(new_devid));
8047 
8048 	/*
8049 	 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
8050 	 * as defined in RFC5661) in XDR.
8051 	 */
8052 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
8053 	    6 * NFSX_UNSIGNED;
8054 	NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
8055 	ds->nfsdev_fileaddrlen = addrlen;
8056 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
8057 	ds->nfsdev_fileaddr = (char *)tl;
8058 	*tl++ = txdr_unsigned(1);		/* One stripe with index 0. */
8059 	*tl++ = 0;
8060 	*tl++ = txdr_unsigned(1);		/* One multipath list */
8061 	*tl++ = txdr_unsigned(1);		/* with one entry in it. */
8062 	/* The netaddr for this one entry. */
8063 	*tl++ = txdr_unsigned(strlen(netprot));
8064 	NFSBCOPY(netprot, tl, strlen(netprot));
8065 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
8066 	*tl++ = txdr_unsigned(strlen(addr));
8067 	NFSBCOPY(addr, tl, strlen(addr));
8068 
8069 	/*
8070 	 * Fill in the flex file addr (actually the ff_device_addr4
8071 	 * as defined for Flexible File Layout) in XDR.
8072 	 */
8073 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
8074 	    14 * NFSX_UNSIGNED;
8075 	ds->nfsdev_flexaddrlen = addrlen;
8076 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
8077 	ds->nfsdev_flexaddr = (char *)tl;
8078 	*tl++ = txdr_unsigned(1);		/* One multipath entry. */
8079 	/* The netaddr for this one entry. */
8080 	*tl++ = txdr_unsigned(strlen(netprot));
8081 	NFSBCOPY(netprot, tl, strlen(netprot));
8082 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
8083 	*tl++ = txdr_unsigned(strlen(addr));
8084 	NFSBCOPY(addr, tl, strlen(addr));
8085 	tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
8086 	*tl++ = txdr_unsigned(2);		/* Two NFS Versions. */
8087 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
8088 	*tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */
8089 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
8090 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
8091 	*tl++ = newnfs_true;			/* Tightly coupled. */
8092 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
8093 	*tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
8094 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
8095 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
8096 	*tl = newnfs_true;			/* Tightly coupled. */
8097 
8098 	ds->nfsdev_hostnamelen = strlen(dnshost);
8099 	ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
8100 	    M_WAITOK);
8101 	NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
8102 }
8103 
8104 /*
8105  * Create the device id list.
8106  * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
8107  * is misconfigured.
8108  */
8109 int
8110 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
8111 {
8112 	struct nfsdevice *ds;
8113 	char *addrp, *dnshostp, *dspathp, *mdspathp;
8114 	int error, i;
8115 
8116 	addrp = args->addr;
8117 	dnshostp = args->dnshost;
8118 	dspathp = args->dspath;
8119 	mdspathp = args->mdspath;
8120 	nfsrv_maxpnfsmirror = args->mirrorcnt;
8121 	if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
8122 	    mdspathp == NULL)
8123 		return (0);
8124 
8125 	/*
8126 	 * Loop around for each nul-terminated string in args->addr,
8127 	 * args->dnshost, args->dnspath and args->mdspath.
8128 	 */
8129 	while (addrp < (args->addr + args->addrlen) &&
8130 	    dnshostp < (args->dnshost + args->dnshostlen) &&
8131 	    dspathp < (args->dspath + args->dspathlen) &&
8132 	    mdspathp < (args->mdspath + args->mdspathlen)) {
8133 		error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
8134 		if (error != 0) {
8135 			/* Free all DS servers. */
8136 			nfsrv_freealldevids();
8137 			nfsrv_devidcnt = 0;
8138 			return (ENXIO);
8139 		}
8140 		nfsrv_allocdevid(ds, addrp, dnshostp);
8141 		addrp += (strlen(addrp) + 1);
8142 		dnshostp += (strlen(dnshostp) + 1);
8143 		dspathp += (strlen(dspathp) + 1);
8144 		mdspathp += (strlen(mdspathp) + 1);
8145 	}
8146 	if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
8147 		/* Free all DS servers. */
8148 		nfsrv_freealldevids();
8149 		nfsrv_devidcnt = 0;
8150 		nfsrv_maxpnfsmirror = 1;
8151 		return (ENXIO);
8152 	}
8153 	/* We can fail at most one less DS than the mirror level. */
8154 	nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
8155 
8156 	/*
8157 	 * Allocate the nfslayout hash table now, since this is a pNFS server.
8158 	 * Make it 1% of the high water mark and at least 100.
8159 	 */
8160 	if (nfslayouthash == NULL) {
8161 		nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
8162 		if (nfsrv_layouthashsize < 100)
8163 			nfsrv_layouthashsize = 100;
8164 		nfslayouthash = mallocarray(nfsrv_layouthashsize,
8165 		    sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
8166 		    M_ZERO);
8167 		for (i = 0; i < nfsrv_layouthashsize; i++) {
8168 			mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
8169 			TAILQ_INIT(&nfslayouthash[i].list);
8170 		}
8171 	}
8172 	return (0);
8173 }
8174 
8175 /*
8176  * Free all device ids.
8177  */
8178 static void
8179 nfsrv_freealldevids(void)
8180 {
8181 	struct nfsdevice *ds, *nds;
8182 
8183 	TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
8184 		nfsrv_freedevid(ds);
8185 }
8186 
8187 /*
8188  * Check to see if there is a Read/Write Layout plus either:
8189  * - A Write Delegation
8190  * or
8191  * - An Open with Write_access.
8192  * Return 1 if this is the case and 0 otherwise.
8193  * This function is used by nfsrv_proxyds() to decide if doing a Proxy
8194  * Getattr RPC to the Data Server (DS) is necessary.
8195  */
8196 #define	NFSCLIDVECSIZE	6
8197 int
8198 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p)
8199 {
8200 	fhandle_t fh, *tfhp;
8201 	struct nfsstate *stp;
8202 	struct nfslayout *lyp;
8203 	struct nfslayouthash *lhyp;
8204 	struct nfslockhashhead *hp;
8205 	struct nfslockfile *lfp;
8206 	nfsquad_t clid[NFSCLIDVECSIZE];
8207 	int clidcnt, ret;
8208 
8209 	ret = nfsvno_getfh(vp, &fh, p);
8210 	if (ret != 0)
8211 		return (0);
8212 
8213 	/* First check for a Read/Write Layout. */
8214 	clidcnt = 0;
8215 	lhyp = NFSLAYOUTHASH(&fh);
8216 	NFSLOCKLAYOUT(lhyp);
8217 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8218 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8219 		    ((lyp->lay_flags & NFSLAY_RW) != 0 ||
8220 		     ((lyp->lay_flags & NFSLAY_READ) != 0 &&
8221 		      nfsrv_pnfsatime != 0))) {
8222 			if (clidcnt < NFSCLIDVECSIZE)
8223 				clid[clidcnt].qval = lyp->lay_clientid.qval;
8224 			clidcnt++;
8225 		}
8226 	}
8227 	NFSUNLOCKLAYOUT(lhyp);
8228 	if (clidcnt == 0) {
8229 		/* None found, so return 0. */
8230 		return (0);
8231 	}
8232 
8233 	/* Get the nfslockfile for this fh. */
8234 	NFSLOCKSTATE();
8235 	hp = NFSLOCKHASH(&fh);
8236 	LIST_FOREACH(lfp, hp, lf_hash) {
8237 		tfhp = &lfp->lf_fh;
8238 		if (NFSVNO_CMPFH(&fh, tfhp))
8239 			break;
8240 	}
8241 	if (lfp == NULL) {
8242 		/* None found, so return 0. */
8243 		NFSUNLOCKSTATE();
8244 		return (0);
8245 	}
8246 
8247 	/* Now, look for a Write delegation for this clientid. */
8248 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
8249 		if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8250 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8251 			break;
8252 	}
8253 	if (stp != NULL) {
8254 		/* Found one, so return 1. */
8255 		NFSUNLOCKSTATE();
8256 		return (1);
8257 	}
8258 
8259 	/* No Write delegation, so look for an Open with Write_access. */
8260 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
8261 		KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
8262 		    ("nfsrv_checkdsattr: Non-open in Open list\n"));
8263 		if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
8264 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8265 			break;
8266 	}
8267 	NFSUNLOCKSTATE();
8268 	if (stp != NULL)
8269 		return (1);
8270 	return (0);
8271 }
8272 
8273 /*
8274  * Look for a matching clientid in the vector. Return 1 if one might match.
8275  */
8276 static int
8277 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
8278 {
8279 	int i;
8280 
8281 	/* If too many for the vector, return 1 since there might be a match. */
8282 	if (clidcnt > NFSCLIDVECSIZE)
8283 		return (1);
8284 
8285 	for (i = 0; i < clidcnt; i++)
8286 		if (clidvec[i].qval == clid.qval)
8287 			return (1);
8288 	return (0);
8289 }
8290 
8291 /*
8292  * Check the don't list for "vp" and see if issuing an rw layout is allowed.
8293  * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
8294  */
8295 static int
8296 nfsrv_dontlayout(fhandle_t *fhp)
8297 {
8298 	struct nfsdontlist *mrp;
8299 	int ret;
8300 
8301 	if (nfsrv_dontlistlen == 0)
8302 		return (0);
8303 	ret = 0;
8304 	NFSDDONTLISTLOCK();
8305 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8306 		if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
8307 		    (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
8308 			ret = 1;
8309 			break;
8310 		}
8311 	}
8312 	NFSDDONTLISTUNLOCK();
8313 	return (ret);
8314 }
8315 
8316 #define	PNFSDS_COPYSIZ	65536
8317 /*
8318  * Create a new file on a DS and copy the contents of an extant DS file to it.
8319  * This can be used for recovery of a DS file onto a recovered DS.
8320  * The steps are:
8321  * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
8322  * - Disable issuing of read/write layouts for the file via the nfsdontlist,
8323  *   so that they will be disabled after the MDS file's vnode is unlocked.
8324  * - Set up the nfsrv_recalllist so that recall of read/write layouts can
8325  *   be done.
8326  * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
8327  *   writes, LayoutCommits and LayoutReturns for the file when completing the
8328  *   LayoutReturn requested by the LayoutRecall callback.
8329  * - Issue a LayoutRecall callback for all read/write layouts and wait for
8330  *   them to be returned. (If the LayoutRecall callback replies
8331  *   NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
8332  * - Exclusively lock the MDS file's vnode.  This ensures that no proxied
8333  *   writes are in progress or can occur during the DS file copy.
8334  *   It also blocks Setattr operations.
8335  * - Create the file on the recovered mirror.
8336  * - Copy the file from the operational DS.
8337  * - Copy any ACL from the MDS file to the new DS file.
8338  * - Set the modify time of the new DS file to that of the MDS file.
8339  * - Update the extended attribute for the MDS file.
8340  * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
8341  * - The caller will unlock the MDS file's vnode allowing operations
8342  *   to continue normally, since it is now on the mirror again.
8343  */
8344 int
8345 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
8346     struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
8347     struct ucred *cred, NFSPROC_T *p)
8348 {
8349 	struct nfsdontlist *mrp, *nmrp;
8350 	struct nfslayouthash *lhyp;
8351 	struct nfslayout *lyp, *nlyp;
8352 	struct nfslayouthead thl;
8353 	struct mount *mp, *tvmp;
8354 	struct acl *aclp;
8355 	struct vattr va;
8356 	struct timespec mtime;
8357 	fhandle_t fh;
8358 	vnode_t tvp;
8359 	off_t rdpos, wrpos;
8360 	ssize_t aresid;
8361 	char *dat;
8362 	int didprintf, ret, retacl, xfer;
8363 
8364 	ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
8365 	ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
8366 	/*
8367 	 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
8368 	 * so that no more RW layouts will get issued.
8369 	 */
8370 	ret = nfsvno_getfh(vp, &fh, p);
8371 	if (ret != 0) {
8372 		NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8373 		return (ret);
8374 	}
8375 	nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8376 	nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8377 	NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8378 	NFSDDONTLISTLOCK();
8379 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8380 		if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8381 			break;
8382 	}
8383 	if (mrp == NULL) {
8384 		LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8385 		mrp = nmrp;
8386 		nmrp = NULL;
8387 		nfsrv_dontlistlen++;
8388 		NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8389 	} else {
8390 		NFSDDONTLISTUNLOCK();
8391 		free(nmrp, M_NFSDSTATE);
8392 		NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8393 		return (ENXIO);
8394 	}
8395 	NFSDDONTLISTUNLOCK();
8396 
8397 	/*
8398 	 * Search for all RW layouts for this file.  Move them to the
8399 	 * recall list, so they can be recalled and their return noted.
8400 	 */
8401 	lhyp = NFSLAYOUTHASH(&fh);
8402 	NFSDRECALLLOCK();
8403 	NFSLOCKLAYOUT(lhyp);
8404 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8405 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8406 		    (lyp->lay_flags & NFSLAY_RW) != 0) {
8407 			TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8408 			TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8409 			lyp->lay_trycnt = 0;
8410 		}
8411 	}
8412 	NFSUNLOCKLAYOUT(lhyp);
8413 	NFSDRECALLUNLOCK();
8414 
8415 	ret = 0;
8416 	mp = tvmp = NULL;
8417 	didprintf = 0;
8418 	TAILQ_INIT(&thl);
8419 	/* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8420 	NFSVOPUNLOCK(vp);
8421 	/* Now, do a recall for all layouts not yet recalled. */
8422 tryagain:
8423 	NFSDRECALLLOCK();
8424 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8425 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8426 		    (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8427 			lyp->lay_flags |= NFSLAY_RECALL;
8428 			/*
8429 			 * The layout stateid.seqid needs to be incremented
8430 			 * before doing a LAYOUT_RECALL callback.
8431 			 */
8432 			if (++lyp->lay_stateid.seqid == 0)
8433 				lyp->lay_stateid.seqid = 1;
8434 			NFSDRECALLUNLOCK();
8435 			nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8436 			    &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8437 			NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8438 			goto tryagain;
8439 		}
8440 	}
8441 
8442 	/* Now wait for them to be returned. */
8443 tryagain2:
8444 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8445 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8446 			if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8447 				TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8448 				    lay_list);
8449 				TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8450 				NFSD_DEBUG(4,
8451 				    "nfsrv_copymr: layout returned\n");
8452 			} else {
8453 				lyp->lay_trycnt++;
8454 				ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8455 				    PVFS | PCATCH, "nfsmrl", hz);
8456 				NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8457 				    ret);
8458 				if (ret == EINTR || ret == ERESTART)
8459 					break;
8460 				if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8461 					/*
8462 					 * Give up after 60sec and return
8463 					 * ENXIO, failing the copymr.
8464 					 * This layout will remain on the
8465 					 * recalllist.  It can only be cleared
8466 					 * by restarting the nfsd.
8467 					 * This seems the safe way to handle
8468 					 * it, since it cannot be safely copied
8469 					 * with an outstanding RW layout.
8470 					 */
8471 					if (lyp->lay_trycnt >= 60) {
8472 						ret = ENXIO;
8473 						break;
8474 					}
8475 					if (didprintf == 0) {
8476 						printf("nfsrv_copymr: layout "
8477 						    "not returned\n");
8478 						didprintf = 1;
8479 					}
8480 				}
8481 			}
8482 			goto tryagain2;
8483 		}
8484 	}
8485 	NFSDRECALLUNLOCK();
8486 	/* We can now get rid of the layouts that have been returned. */
8487 	TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8488 		nfsrv_freelayout(&thl, lyp);
8489 
8490 	/*
8491 	 * Do the vn_start_write() calls here, before the MDS vnode is
8492 	 * locked and the tvp is created (locked) in the NFS file system
8493 	 * that dvp is in.
8494 	 * For tvmp, this probably isn't necessary, since it will be an
8495 	 * NFS mount and they are not suspendable at this time.
8496 	 */
8497 	if (ret == 0)
8498 		ret = vn_start_write(vp, &mp, V_WAIT | V_PCATCH);
8499 	if (ret == 0) {
8500 		tvmp = dvp->v_mount;
8501 		ret = vn_start_write(NULL, &tvmp, V_WAIT | V_PCATCH);
8502 	}
8503 
8504 	/*
8505 	 * LK_EXCLUSIVE lock the MDS vnode, so that any
8506 	 * proxied writes through the MDS will be blocked until we have
8507 	 * completed the copy and update of the extended attributes.
8508 	 * This will also ensure that any attributes and ACL will not be
8509 	 * changed until the copy is complete.
8510 	 */
8511 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8512 	if (ret == 0 && VN_IS_DOOMED(vp)) {
8513 		NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8514 		ret = ESTALE;
8515 	}
8516 
8517 	/* Create the data file on the recovered DS. */
8518 	if (ret == 0)
8519 		ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8520 
8521 	/* Copy the DS file, if created successfully. */
8522 	if (ret == 0) {
8523 		/*
8524 		 * Get any NFSv4 ACL on the MDS file, so that it can be set
8525 		 * on the new DS file.
8526 		 */
8527 		aclp = acl_alloc(M_WAITOK | M_ZERO);
8528 		retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8529 		if (retacl != 0 && retacl != ENOATTR)
8530 			NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8531 		dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8532 		/* Malloc a block of 0s used to check for holes. */
8533 		if (nfsrv_zeropnfsdat == NULL)
8534 			nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8535 			    M_WAITOK | M_ZERO);
8536 		rdpos = wrpos = 0;
8537 		ret = VOP_GETATTR(fvp, &va, cred);
8538 		aresid = 0;
8539 		while (ret == 0 && aresid == 0) {
8540 			ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8541 			    rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8542 			    &aresid, p);
8543 			xfer = PNFSDS_COPYSIZ - aresid;
8544 			if (ret == 0 && xfer > 0) {
8545 				rdpos += xfer;
8546 				/*
8547 				 * Skip the write for holes, except for the
8548 				 * last block.
8549 				 */
8550 				if (xfer < PNFSDS_COPYSIZ || rdpos ==
8551 				    va.va_size || NFSBCMP(dat,
8552 				    nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8553 					ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8554 					    wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8555 					    cred, NULL, NULL, p);
8556 				if (ret == 0)
8557 					wrpos += xfer;
8558 			}
8559 		}
8560 
8561 		/* If there is an ACL and the copy succeeded, set the ACL. */
8562 		if (ret == 0 && retacl == 0) {
8563 			ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8564 			/*
8565 			 * Don't consider these as errors, since VOP_GETACL()
8566 			 * can return an ACL when they are not actually
8567 			 * supported.  For example, for UFS, VOP_GETACL()
8568 			 * will return a trivial ACL based on the uid/gid/mode
8569 			 * when there is no ACL on the file.
8570 			 * This case should be recognized as a trivial ACL
8571 			 * by UFS's VOP_SETACL() and succeed, but...
8572 			 */
8573 			if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8574 				ret = 0;
8575 		}
8576 
8577 		if (ret == 0)
8578 			ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8579 
8580 		/* Set the DS data file's modify time that of the MDS file. */
8581 		if (ret == 0)
8582 			ret = VOP_GETATTR(vp, &va, cred);
8583 		if (ret == 0) {
8584 			mtime = va.va_mtime;
8585 			VATTR_NULL(&va);
8586 			va.va_mtime = mtime;
8587 			ret = VOP_SETATTR(tvp, &va, cred);
8588 		}
8589 
8590 		vput(tvp);
8591 		acl_free(aclp);
8592 		free(dat, M_TEMP);
8593 	}
8594 	if (tvmp != NULL)
8595 		vn_finished_write(tvmp);
8596 
8597 	/* Update the extended attributes for the newly created DS file. */
8598 	if (ret == 0)
8599 		ret = vn_extattr_set(vp, IO_NODELOCKED,
8600 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8601 		    sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8602 	if (mp != NULL)
8603 		vn_finished_write(mp);
8604 
8605 	/* Get rid of the dontlist entry, so that Layouts can be issued. */
8606 	NFSDDONTLISTLOCK();
8607 	LIST_REMOVE(mrp, nfsmr_list);
8608 	NFSDDONTLISTUNLOCK();
8609 	free(mrp, M_NFSDSTATE);
8610 	return (ret);
8611 }
8612 
8613 /*
8614  * Create a data storage file on the recovered DS.
8615  */
8616 static int
8617 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8618     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8619     vnode_t *tvpp)
8620 {
8621 	struct vattr va, nva;
8622 	int error;
8623 
8624 	/* Make data file name based on FH. */
8625 	error = VOP_GETATTR(vp, &va, cred);
8626 	if (error == 0) {
8627 		/* Set the attributes for "vp" to Setattr the DS vp. */
8628 		VATTR_NULL(&nva);
8629 		nva.va_uid = va.va_uid;
8630 		nva.va_gid = va.va_gid;
8631 		nva.va_mode = va.va_mode;
8632 		nva.va_size = 0;
8633 		VATTR_NULL(&va);
8634 		va.va_type = VREG;
8635 		va.va_mode = nva.va_mode;
8636 		NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8637 		error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8638 		    pf->dsf_filename, cred, p, tvpp);
8639 	}
8640 	return (error);
8641 }
8642 
8643 /*
8644  * Look up the MDS file shared locked, and then get the extended attribute
8645  * to find the extant DS file to be copied to the new mirror.
8646  * If successful, *vpp is set to the MDS file's vp and *nvpp is
8647  * set to a DS data file for the MDS file, both exclusively locked.
8648  * The "buf" argument has the pnfsdsfile structure from the MDS file
8649  * in it and buflen is set to its length.
8650  */
8651 int
8652 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8653     int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8654     struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8655     struct nfsdevice **fdsp)
8656 {
8657 	struct nameidata nd;
8658 	struct vnode *vp, *curvp;
8659 	struct pnfsdsfile *pf;
8660 	struct nfsmount *nmp, *curnmp;
8661 	int dsdir, error, mirrorcnt, ippos;
8662 
8663 	vp = NULL;
8664 	curvp = NULL;
8665 	curnmp = NULL;
8666 	*dsp = NULL;
8667 	*fdsp = NULL;
8668 	if (dspathp == NULL && curdspathp != NULL)
8669 		return (EPERM);
8670 
8671 	/*
8672 	 * Look up the MDS file shared locked.  The lock will be upgraded
8673 	 * to an exclusive lock after any rw layouts have been returned.
8674 	 */
8675 	NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8676 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8677 	    mdspathp);
8678 	error = namei(&nd);
8679 	NFSD_DEBUG(4, "lookup=%d\n", error);
8680 	if (error != 0)
8681 		return (error);
8682 	if (nd.ni_vp->v_type != VREG) {
8683 		vput(nd.ni_vp);
8684 		NFSD_DEBUG(4, "mdspath not reg\n");
8685 		return (EISDIR);
8686 	}
8687 	vp = nd.ni_vp;
8688 
8689 	if (curdspathp != NULL) {
8690 		/*
8691 		 * Look up the current DS path and find the nfsdev structure for
8692 		 * it.
8693 		 */
8694 		NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8695 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8696 		    UIO_SYSSPACE, curdspathp);
8697 		error = namei(&nd);
8698 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8699 		if (error != 0) {
8700 			vput(vp);
8701 			return (error);
8702 		}
8703 		if (nd.ni_vp->v_type != VDIR) {
8704 			vput(nd.ni_vp);
8705 			vput(vp);
8706 			NFSD_DEBUG(4, "curdspath not dir\n");
8707 			return (ENOTDIR);
8708 		}
8709 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8710 			vput(nd.ni_vp);
8711 			vput(vp);
8712 			NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8713 			return (ENXIO);
8714 		}
8715 		curnmp = VFSTONFS(nd.ni_vp->v_mount);
8716 
8717 		/* Search the nfsdev list for a match. */
8718 		NFSDDSLOCK();
8719 		*fdsp = nfsv4_findmirror(curnmp);
8720 		NFSDDSUNLOCK();
8721 		if (*fdsp == NULL)
8722 			curnmp = NULL;
8723 		if (curnmp == NULL) {
8724 			vput(nd.ni_vp);
8725 			vput(vp);
8726 			NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8727 			return (ENXIO);
8728 		}
8729 		curvp = nd.ni_vp;
8730 	}
8731 
8732 	if (dspathp != NULL) {
8733 		/* Look up the nfsdev path and find the nfsdev structure. */
8734 		NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8735 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8736 		    UIO_SYSSPACE, dspathp);
8737 		error = namei(&nd);
8738 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8739 		if (error != 0) {
8740 			vput(vp);
8741 			if (curvp != NULL)
8742 				vput(curvp);
8743 			return (error);
8744 		}
8745 		if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8746 			vput(nd.ni_vp);
8747 			vput(vp);
8748 			if (curvp != NULL)
8749 				vput(curvp);
8750 			NFSD_DEBUG(4, "dspath not dir\n");
8751 			if (nd.ni_vp == curvp)
8752 				return (EPERM);
8753 			return (ENOTDIR);
8754 		}
8755 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8756 			vput(nd.ni_vp);
8757 			vput(vp);
8758 			if (curvp != NULL)
8759 				vput(curvp);
8760 			NFSD_DEBUG(4, "dspath not an NFS mount\n");
8761 			return (ENXIO);
8762 		}
8763 		nmp = VFSTONFS(nd.ni_vp->v_mount);
8764 
8765 		/*
8766 		 * Search the nfsdevice list for a match.  If curnmp == NULL,
8767 		 * this is a recovery and there must be a mirror.
8768 		 */
8769 		NFSDDSLOCK();
8770 		if (curnmp == NULL)
8771 			*dsp = nfsrv_findmirroredds(nmp);
8772 		else
8773 			*dsp = nfsv4_findmirror(nmp);
8774 		NFSDDSUNLOCK();
8775 		if (*dsp == NULL) {
8776 			vput(nd.ni_vp);
8777 			vput(vp);
8778 			if (curvp != NULL)
8779 				vput(curvp);
8780 			NFSD_DEBUG(4, "mdscopymr: no ds\n");
8781 			return (ENXIO);
8782 		}
8783 	} else {
8784 		nd.ni_vp = NULL;
8785 		nmp = NULL;
8786 	}
8787 
8788 	/*
8789 	 * Get a vp for an available DS data file using the extended
8790 	 * attribute on the MDS file.
8791 	 * If there is a valid entry for the new DS in the extended attribute
8792 	 * on the MDS file (as checked via the nmp argument),
8793 	 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8794 	 */
8795 	error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p,
8796 	    NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8797 	if (curvp != NULL)
8798 		vput(curvp);
8799 	if (nd.ni_vp == NULL) {
8800 		if (error == 0 && nmp != NULL) {
8801 			/* Search the nfsdev list for a match. */
8802 			NFSDDSLOCK();
8803 			*dsp = nfsrv_findmirroredds(nmp);
8804 			NFSDDSUNLOCK();
8805 		}
8806 		if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8807 			if (nvpp != NULL && *nvpp != NULL) {
8808 				vput(*nvpp);
8809 				*nvpp = NULL;
8810 			}
8811 			error = ENXIO;
8812 		}
8813 	} else
8814 		vput(nd.ni_vp);
8815 
8816 	/*
8817 	 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8818 	 * and is only allowed if there is a 0.0.0.0 IP address entry.
8819 	 * When curdspathp != NULL, the ippos will be set to that entry.
8820 	 */
8821 	if (error == 0 && dspathp != NULL && ippos == -1) {
8822 		if (nvpp != NULL && *nvpp != NULL) {
8823 			vput(*nvpp);
8824 			*nvpp = NULL;
8825 		}
8826 		error = ENXIO;
8827 	}
8828 	if (error == 0) {
8829 		*vpp = vp;
8830 
8831 		pf = (struct pnfsdsfile *)buf;
8832 		if (ippos == -1) {
8833 			/* If no zeroip pnfsdsfile, add one. */
8834 			ippos = *buflenp / sizeof(*pf);
8835 			*buflenp += sizeof(*pf);
8836 			pf += ippos;
8837 			pf->dsf_dir = dsdir;
8838 			strlcpy(pf->dsf_filename, fname,
8839 			    sizeof(pf->dsf_filename));
8840 		} else
8841 			pf += ippos;
8842 		*pfp = pf;
8843 	} else
8844 		vput(vp);
8845 	return (error);
8846 }
8847 
8848 /*
8849  * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8850  * Return one if found, NULL otherwise.
8851  */
8852 static struct nfsdevice *
8853 nfsrv_findmirroredds(struct nfsmount *nmp)
8854 {
8855 	struct nfsdevice *ds, *fndds;
8856 	int fndmirror;
8857 
8858 	mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8859 	/*
8860 	 * Search the DS server list for a match with nmp.
8861 	 * Remove the DS entry if found and there is a mirror.
8862 	 */
8863 	fndds = NULL;
8864 	fndmirror = 0;
8865 	if (nfsrv_devidcnt == 0)
8866 		return (fndds);
8867 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8868 		if (ds->nfsdev_nmp == nmp) {
8869 			NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8870 			fndds = ds;
8871 			break;
8872 		}
8873 	}
8874 	if (fndds == NULL)
8875 		return (fndds);
8876 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8877 		fndmirror = 1;
8878 	else if (fndds->nfsdev_mdsisset != 0) {
8879 		/* For the fsid is set case, search for a mirror. */
8880 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8881 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
8882 			    ds->nfsdev_mdsisset != 0 &&
8883 			    fsidcmp(&ds->nfsdev_mdsfsid,
8884 			    &fndds->nfsdev_mdsfsid) == 0) {
8885 				fndmirror = 1;
8886 				break;
8887 			}
8888 		}
8889 	}
8890 	if (fndmirror == 0) {
8891 		NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");
8892 		return (NULL);
8893 	}
8894 	return (fndds);
8895 }
8896 
8897 /*
8898  * Mark the appropriate devid and all associated layout as "out of space".
8899  */
8900 void
8901 nfsrv_marknospc(char *devid, bool setit)
8902 {
8903 	struct nfsdevice *ds;
8904 	struct nfslayout *lyp;
8905 	struct nfslayouthash *lhyp;
8906 	int i;
8907 
8908 	NFSDDSLOCK();
8909 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8910 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0) {
8911 			NFSD_DEBUG(1, "nfsrv_marknospc: devid %d\n", setit);
8912 			ds->nfsdev_nospc = setit;
8913 		}
8914 	}
8915 	NFSDDSUNLOCK();
8916 
8917 	for (i = 0; i < nfsrv_layouthashsize; i++) {
8918 		lhyp = &nfslayouthash[i];
8919 		NFSLOCKLAYOUT(lhyp);
8920 		TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8921 			if (NFSBCMP(lyp->lay_deviceid, devid,
8922 			    NFSX_V4DEVICEID) == 0) {
8923 				NFSD_DEBUG(1, "nfsrv_marknospc: layout %d\n",
8924 				    setit);
8925 				if (setit)
8926 					lyp->lay_flags |= NFSLAY_NOSPC;
8927 				else
8928 					lyp->lay_flags &= ~NFSLAY_NOSPC;
8929 			}
8930 		}
8931 		NFSUNLOCKLAYOUT(lhyp);
8932 	}
8933 }
8934 
8935 /*
8936  * Check to see if SP4_MACH_CRED is in use and, if it is, check that the
8937  * correct machine credential is being used.
8938  */
8939 static int
8940 nfsrv_checkmachcred(int op, struct nfsrv_descript *nd, struct nfsclient *clp)
8941 {
8942 
8943 	if ((clp->lc_flags & LCL_MACHCRED) == 0 ||
8944 	    !NFSISSET_OPBIT(&clp->lc_mustops, op))
8945 		return (0);
8946 	KASSERT((nd->nd_flag & ND_NFSV41) != 0,
8947 	    ("nfsrv_checkmachcred: MachCred for NFSv4.0"));
8948 	if ((nd->nd_flag & (ND_GSSINTEGRITY | ND_GSSPRIVACY)) != 0 &&
8949 	    nd->nd_princlen == clp->lc_namelen &&
8950 	    !NFSBCMP(nd->nd_principal, clp->lc_name, nd->nd_princlen))
8951 		return (0);
8952 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
8953 }
8954