xref: /freebsd/sys/fs/nfsclient/nfs_clstate.c (revision aa7208254996a66fb0b363bb696fe65d81335d81)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2009 Rick Macklem, University of Guelph
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 /*
34  * These functions implement the client side state handling for NFSv4.
35  * NFSv4 state handling:
36  * - A lockowner is used to determine lock contention, so it
37  *   corresponds directly to a Posix pid. (1 to 1 mapping)
38  * - The correct granularity of an OpenOwner is not nearly so
39  *   obvious. An OpenOwner does the following:
40  *   - provides a serial sequencing of Open/Close/Lock-with-new-lockowner
41  *   - is used to check for Open/Share contention (not applicable to
42  *     this client, since all Opens are Deny_None)
43  *   As such, I considered both extreme.
44  *   1 OpenOwner per ClientID - Simple to manage, but fully serializes
45  *   all Open, Close and Lock (with a new lockowner) Ops.
46  *   1 OpenOwner for each Open - This one results in an OpenConfirm for
47  *   every Open, for most servers.
48  *   So, I chose to use the same mapping as I did for LockOwnwers.
49  *   The main concern here is that you can end up with multiple Opens
50  *   for the same File Handle, but on different OpenOwners (opens
51  *   inherited from parents, grandparents...) and you do not know
52  *   which of these the vnodeop close applies to. This is handled by
53  *   delaying the Close Op(s) until all of the Opens have been closed.
54  *   (It is not yet obvious if this is the correct granularity.)
55  * - How the code handles serialization:
56  *   - For the ClientId, it uses an exclusive lock while getting its
57  *     SetClientId and during recovery. Otherwise, it uses a shared
58  *     lock via a reference count.
59  *   - For the rest of the data structures, it uses an SMP mutex
60  *     (once the nfs client is SMP safe) and doesn't sleep while
61  *     manipulating the linked lists.
62  *   - The serialization of Open/Close/Lock/LockU falls out in the
63  *     "wash", since OpenOwners and LockOwners are both mapped from
64  *     Posix pid. In other words, there is only one Posix pid using
65  *     any given owner, so that owner is serialized. (If you change
66  *     the granularity of the OpenOwner, then code must be added to
67  *     serialize Ops on the OpenOwner.)
68  * - When to get rid of OpenOwners and LockOwners.
69  *   - The function nfscl_cleanup_common() is executed after a process exits.
70  *     It goes through the client list looking for all Open and Lock Owners.
71  *     When one is found, it is marked "defunct" or in the case of
72  *     an OpenOwner without any Opens, freed.
73  *     The renew thread scans for defunct Owners and gets rid of them,
74  *     if it can. The LockOwners will also be deleted when the
75  *     associated Open is closed.
76  *   - If the LockU or Close Op(s) fail during close in a way
77  *     that could be recovered upon retry, they are relinked to the
78  *     ClientId's defunct open list and retried by the renew thread
79  *     until they succeed or an unmount/recovery occurs.
80  *     (Since we are done with them, they do not need to be recovered.)
81  */
82 
83 #include <fs/nfs/nfsport.h>
84 
85 /*
86  * Global variables
87  */
88 extern struct nfsstatsv1 nfsstatsv1;
89 extern struct nfsreqhead nfsd_reqq;
90 extern u_int32_t newnfs_false, newnfs_true;
91 extern int nfscl_debuglevel;
92 extern int nfscl_enablecallb;
93 extern int nfs_numnfscbd;
94 NFSREQSPINLOCK;
95 NFSCLSTATEMUTEX;
96 int nfscl_inited = 0;
97 struct nfsclhead nfsclhead;	/* Head of clientid list */
98 int nfscl_deleghighwater = NFSCLDELEGHIGHWATER;
99 int nfscl_layouthighwater = NFSCLLAYOUTHIGHWATER;
100 
101 static int nfscl_delegcnt = 0;
102 static int nfscl_layoutcnt = 0;
103 static int nfscl_getopen(struct nfsclownerhead *, struct nfsclopenhash *,
104     u_int8_t *, int, u_int8_t *, u_int8_t *, u_int32_t,
105     struct nfscllockowner **, struct nfsclopen **);
106 static bool nfscl_checkown(struct nfsclowner *, struct nfsclopen *, uint8_t *,
107     uint8_t *, struct nfscllockowner **, struct nfsclopen **,
108     struct nfsclopen **);
109 static void nfscl_clrelease(struct nfsclclient *);
110 static void nfscl_unlinkopen(struct nfsclopen *);
111 static void nfscl_cleanclient(struct nfsclclient *);
112 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
113     struct ucred *, NFSPROC_T *);
114 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
115     struct nfsmount *, struct ucred *, NFSPROC_T *);
116 static void nfscl_recover(struct nfsclclient *, bool *, struct ucred *,
117     NFSPROC_T *);
118 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
119     struct nfscllock *, int);
120 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
121     struct nfscllock **, int);
122 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *,
123     struct nfscldeleghead *);
124 static u_int32_t nfscl_nextcbident(void);
125 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
126 static struct nfsclclient *nfscl_getclnt(u_int32_t);
127 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
128 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
129     int);
130 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
131     int, struct nfsclrecalllayout **, struct nfscllayout **);
132 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
133 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
134     int);
135 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
136 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
137     u_int8_t *, struct nfscllock **);
138 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
139 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
140     struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
141 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
142     struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
143     struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
144 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
145     struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
146     struct nfscldeleg *, struct ucred *, NFSPROC_T *);
147 static void nfscl_totalrecall(struct nfsclclient *);
148 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
149     struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
150 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
151     u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
152     struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
153 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
154     int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
155     struct ucred *, NFSPROC_T *);
156 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
157     struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
158 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *,
159     bool);
160 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
161 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
162 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
163     struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int,
164     vnode_t *);
165 static void nfscl_freeopenowner(struct nfsclowner *, int);
166 static void nfscl_cleandeleg(struct nfscldeleg *);
167 static void nfscl_emptylockowner(struct nfscllockowner *,
168     struct nfscllockownerfhhead *);
169 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
170     struct nfsclflayouthead *);
171 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
172     uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
173 static int nfscl_seq(uint32_t, uint32_t);
174 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
175     struct ucred *, NFSPROC_T *);
176 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
177     struct ucred *, NFSPROC_T *);
178 
179 static short nfscberr_null[] = {
180 	0,
181 	0,
182 };
183 
184 static short nfscberr_getattr[] = {
185 	NFSERR_RESOURCE,
186 	NFSERR_BADHANDLE,
187 	NFSERR_BADXDR,
188 	NFSERR_RESOURCE,
189 	NFSERR_SERVERFAULT,
190 	0,
191 };
192 
193 static short nfscberr_recall[] = {
194 	NFSERR_RESOURCE,
195 	NFSERR_BADHANDLE,
196 	NFSERR_BADSTATEID,
197 	NFSERR_BADXDR,
198 	NFSERR_RESOURCE,
199 	NFSERR_SERVERFAULT,
200 	0,
201 };
202 
203 static short *nfscl_cberrmap[] = {
204 	nfscberr_null,
205 	nfscberr_null,
206 	nfscberr_null,
207 	nfscberr_getattr,
208 	nfscberr_recall
209 };
210 
211 #define	NETFAMILY(clp) \
212 		(((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
213 
214 /*
215  * Called for an open operation.
216  * If the nfhp argument is NULL, just get an openowner.
217  */
218 int
219 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
220     struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
221     struct nfsclopen **opp, int *newonep, int *retp, int lockit, bool firstref)
222 {
223 	struct nfsclclient *clp;
224 	struct nfsclowner *owp, *nowp;
225 	struct nfsclopen *op = NULL, *nop = NULL;
226 	struct nfscldeleg *dp;
227 	struct nfsclownerhead *ohp;
228 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
229 	int ret;
230 
231 	if (newonep != NULL)
232 		*newonep = 0;
233 	if (opp != NULL)
234 		*opp = NULL;
235 	if (owpp != NULL)
236 		*owpp = NULL;
237 
238 	/*
239 	 * Might need one or both of these, so MALLOC them now, to
240 	 * avoid a tsleep() in MALLOC later.
241 	 */
242 	nowp = malloc(sizeof (struct nfsclowner),
243 	    M_NFSCLOWNER, M_WAITOK);
244 	if (nfhp != NULL) {
245 	    nop = malloc(sizeof (struct nfsclopen) +
246 		fhlen - 1, M_NFSCLOPEN, M_WAITOK);
247 	    nop->nfso_hash.le_prev = NULL;
248 	}
249 	ret = nfscl_getcl(vp->v_mount, cred, p, false, firstref, &clp);
250 	if (ret != 0) {
251 		free(nowp, M_NFSCLOWNER);
252 		if (nop != NULL)
253 			free(nop, M_NFSCLOPEN);
254 		return (ret);
255 	}
256 
257 	/*
258 	 * Get the Open iff it already exists.
259 	 * If none found, add the new one or return error, depending upon
260 	 * "create".
261 	 */
262 	NFSLOCKCLSTATE();
263 	dp = NULL;
264 	/* First check the delegation list */
265 	if (nfhp != NULL && usedeleg) {
266 		LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
267 			if (dp->nfsdl_fhlen == fhlen &&
268 			    !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
269 				if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
270 				    (dp->nfsdl_flags & NFSCLDL_WRITE))
271 					break;
272 				dp = NULL;
273 				break;
274 			}
275 		}
276 	}
277 
278 	/* For NFSv4.1/4.2 and this option, use a single open_owner. */
279 	if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
280 		nfscl_filllockowner(NULL, own, F_POSIX);
281 	else
282 		nfscl_filllockowner(p->td_proc, own, F_POSIX);
283 	if (dp != NULL)
284 		ohp = &dp->nfsdl_owner;
285 	else
286 		ohp = &clp->nfsc_owner;
287 	/* Now, search for an openowner */
288 	LIST_FOREACH(owp, ohp, nfsow_list) {
289 		if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
290 			break;
291 	}
292 
293 	/*
294 	 * Create a new open, as required.
295 	 */
296 	nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
297 	    cred, newonep);
298 
299 	/*
300 	 * Now, check the mode on the open and return the appropriate
301 	 * value.
302 	 */
303 	if (retp != NULL) {
304 		if (nfhp != NULL && dp != NULL && nop == NULL)
305 			/* new local open on delegation */
306 			*retp = NFSCLOPEN_SETCRED;
307 		else
308 			*retp = NFSCLOPEN_OK;
309 	}
310 	if (op != NULL && (amode & ~(op->nfso_mode))) {
311 		op->nfso_mode |= amode;
312 		if (retp != NULL && dp == NULL)
313 			*retp = NFSCLOPEN_DOOPEN;
314 	}
315 
316 	/*
317 	 * Serialize modifications to the open owner for multiple threads
318 	 * within the same process using a read/write sleep lock.
319 	 * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
320 	 * by acquiring a shared lock.  The close operations still use an
321 	 * exclusive lock for this case.
322 	 */
323 	if (lockit != 0) {
324 		if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount))) {
325 			/*
326 			 * Get a shared lock on the OpenOwner, but first
327 			 * wait for any pending exclusive lock, so that the
328 			 * exclusive locker gets priority.
329 			 */
330 			nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
331 			    NFSCLSTATEMUTEXPTR, NULL);
332 			nfsv4_getref(&owp->nfsow_rwlock, NULL,
333 			    NFSCLSTATEMUTEXPTR, NULL);
334 		} else
335 			nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
336 	}
337 	NFSUNLOCKCLSTATE();
338 	if (nowp != NULL)
339 		free(nowp, M_NFSCLOWNER);
340 	if (nop != NULL)
341 		free(nop, M_NFSCLOPEN);
342 	if (owpp != NULL)
343 		*owpp = owp;
344 	if (opp != NULL)
345 		*opp = op;
346 	return (0);
347 }
348 
349 /*
350  * Create a new open, as required.
351  */
352 static void
353 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
354     struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
355     struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
356     struct ucred *cred, int *newonep)
357 {
358 	struct nfsclowner *owp = *owpp, *nowp;
359 	struct nfsclopen *op, *nop;
360 
361 	if (nowpp != NULL)
362 		nowp = *nowpp;
363 	else
364 		nowp = NULL;
365 	if (nopp != NULL)
366 		nop = *nopp;
367 	else
368 		nop = NULL;
369 	if (owp == NULL && nowp != NULL) {
370 		NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
371 		LIST_INIT(&nowp->nfsow_open);
372 		nowp->nfsow_clp = clp;
373 		nowp->nfsow_seqid = 0;
374 		nowp->nfsow_defunct = 0;
375 		nfscl_lockinit(&nowp->nfsow_rwlock);
376 		if (dp != NULL) {
377 			nfsstatsv1.cllocalopenowners++;
378 			LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
379 		} else {
380 			nfsstatsv1.clopenowners++;
381 			LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
382 		}
383 		owp = *owpp = nowp;
384 		*nowpp = NULL;
385 		if (newonep != NULL)
386 			*newonep = 1;
387 	}
388 
389 	 /* If an fhp has been specified, create an Open as well. */
390 	if (fhp != NULL) {
391 		/* and look for the correct open, based upon FH */
392 		LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
393 			if (op->nfso_fhlen == fhlen &&
394 			    !NFSBCMP(op->nfso_fh, fhp, fhlen))
395 				break;
396 		}
397 		if (op == NULL && nop != NULL) {
398 			nop->nfso_own = owp;
399 			nop->nfso_mode = 0;
400 			nop->nfso_opencnt = 0;
401 			nop->nfso_posixlock = 1;
402 			nop->nfso_fhlen = fhlen;
403 			NFSBCOPY(fhp, nop->nfso_fh, fhlen);
404 			LIST_INIT(&nop->nfso_lock);
405 			nop->nfso_stateid.seqid = 0;
406 			nop->nfso_stateid.other[0] = 0;
407 			nop->nfso_stateid.other[1] = 0;
408 			nop->nfso_stateid.other[2] = 0;
409 			KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
410 			newnfs_copyincred(cred, &nop->nfso_cred);
411 			if (dp != NULL) {
412 				TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
413 				TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
414 				    nfsdl_list);
415 				dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
416 				nfsstatsv1.cllocalopens++;
417 			} else {
418 				LIST_INSERT_HEAD(NFSCLOPENHASH(clp, fhp, fhlen),
419 				    nop, nfso_hash);
420 				nfsstatsv1.clopens++;
421 			}
422 			LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
423 			*opp = nop;
424 			*nopp = NULL;
425 			if (newonep != NULL)
426 				*newonep = 1;
427 		} else {
428 			*opp = op;
429 		}
430 	}
431 }
432 
433 /*
434  * Called to find/add a delegation to a client.
435  */
436 int
437 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
438     int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
439 {
440 	struct nfscldeleg *dp = *dpp, *tdp;
441 	struct nfsmount *nmp;
442 
443 	KASSERT(mp != NULL, ("nfscl_deleg: mp NULL"));
444 	nmp = VFSTONFS(mp);
445 	/*
446 	 * First, if we have received a Read delegation for a file on a
447 	 * read/write file system, just return it, because they aren't
448 	 * useful, imho.
449 	 */
450 	if (dp != NULL && !NFSMNT_RDONLY(mp) &&
451 	    (dp->nfsdl_flags & NFSCLDL_READ)) {
452 		nfscl_trydelegreturn(dp, cred, nmp, p);
453 		free(dp, M_NFSCLDELEG);
454 		*dpp = NULL;
455 		return (0);
456 	}
457 
458 	/*
459 	 * Since a delegation might be added to the mount,
460 	 * set NFSMNTP_DELEGISSUED now.  If a delegation already
461 	 * exagain ists, setting this flag is harmless.
462 	 */
463 	NFSLOCKMNT(nmp);
464 	nmp->nm_privflag |= NFSMNTP_DELEGISSUED;
465 	NFSUNLOCKMNT(nmp);
466 
467 	/* Look for the correct deleg, based upon FH */
468 	NFSLOCKCLSTATE();
469 	tdp = nfscl_finddeleg(clp, nfhp, fhlen);
470 	if (tdp == NULL) {
471 		if (dp == NULL) {
472 			NFSUNLOCKCLSTATE();
473 			return (NFSERR_BADSTATEID);
474 		}
475 		*dpp = NULL;
476 		TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
477 		LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
478 		    nfsdl_hash);
479 		dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
480 		nfsstatsv1.cldelegates++;
481 		nfscl_delegcnt++;
482 	} else {
483 		/*
484 		 * Delegation already exists, what do we do if a new one??
485 		 */
486 		if (dp != NULL) {
487 			printf("Deleg already exists!\n");
488 			free(dp, M_NFSCLDELEG);
489 			*dpp = NULL;
490 		} else {
491 			*dpp = tdp;
492 		}
493 	}
494 	NFSUNLOCKCLSTATE();
495 	return (0);
496 }
497 
498 /*
499  * Find a delegation for this file handle. Return NULL upon failure.
500  */
501 static struct nfscldeleg *
502 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
503 {
504 	struct nfscldeleg *dp;
505 
506 	LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
507 	    if (dp->nfsdl_fhlen == fhlen &&
508 		!NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
509 		break;
510 	}
511 	return (dp);
512 }
513 
514 /*
515  * Get a stateid for an I/O operation. First, look for an open and iff
516  * found, return either a lockowner stateid or the open stateid.
517  * If no Open is found, just return error and the special stateid of all zeros.
518  */
519 int
520 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
521     int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
522     void **lckpp)
523 {
524 	struct nfsclclient *clp;
525 	struct nfsclopen *op = NULL, *top;
526 	struct nfsclopenhash *oph;
527 	struct nfscllockowner *lp;
528 	struct nfscldeleg *dp;
529 	struct nfsnode *np;
530 	struct nfsmount *nmp;
531 	u_int8_t own[NFSV4CL_LOCKNAMELEN], lockown[NFSV4CL_LOCKNAMELEN];
532 	int error;
533 	bool done;
534 
535 	*lckpp = NULL;
536 	/*
537 	 * Initially, just set the special stateid of all zeros.
538 	 * (Don't do this for a DS, since the special stateid can't be used.)
539 	 */
540 	if (fords == 0) {
541 		stateidp->seqid = 0;
542 		stateidp->other[0] = 0;
543 		stateidp->other[1] = 0;
544 		stateidp->other[2] = 0;
545 	}
546 	if (vnode_vtype(vp) != VREG)
547 		return (EISDIR);
548 	np = VTONFS(vp);
549 	nmp = VFSTONFS(vp->v_mount);
550 
551 	/*
552 	 * For "oneopenown" mounts, first check for a cached open in the
553 	 * NFS vnode, that can be used as a stateid.  This can only be
554 	 * done if no delegations have been issued to the mount and no
555 	 * byte range file locking has been done for the file.
556 	 */
557 	if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && fords == 0) {
558 		NFSLOCKMNT(nmp);
559 		NFSLOCKNODE(np);
560 		if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 &&
561 		    (np->n_flag & NMIGHTBELOCKED) == 0 &&
562 		    np->n_openstateid != NULL) {
563 			stateidp->seqid = 0;
564 			stateidp->other[0] =
565 			    np->n_openstateid->nfso_stateid.other[0];
566 			stateidp->other[1] =
567 			    np->n_openstateid->nfso_stateid.other[1];
568 			stateidp->other[2] =
569 			    np->n_openstateid->nfso_stateid.other[2];
570 			NFSUNLOCKNODE(np);
571 			NFSUNLOCKMNT(nmp);
572 			return (0);
573 		}
574 		NFSUNLOCKNODE(np);
575 		NFSUNLOCKMNT(nmp);
576 	}
577 
578 	NFSLOCKCLSTATE();
579 	clp = nfscl_findcl(nmp);
580 	if (clp == NULL) {
581 		NFSUNLOCKCLSTATE();
582 		return (EACCES);
583 	}
584 
585 	/*
586 	 * Wait for recovery to complete.
587 	 */
588 	while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
589 		(void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
590 		    PZERO, "nfsrecvr", NULL);
591 
592 	/*
593 	 * First, look for a delegation.
594 	 */
595 	LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
596 		if (dp->nfsdl_fhlen == fhlen &&
597 		    !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
598 			if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
599 			    (dp->nfsdl_flags & NFSCLDL_WRITE)) {
600 				if (NFSHASNFSV4N(nmp))
601 					stateidp->seqid = 0;
602 				else
603 					stateidp->seqid =
604 					    dp->nfsdl_stateid.seqid;
605 				stateidp->other[0] = dp->nfsdl_stateid.other[0];
606 				stateidp->other[1] = dp->nfsdl_stateid.other[1];
607 				stateidp->other[2] = dp->nfsdl_stateid.other[2];
608 				if (!(np->n_flag & NDELEGRECALL)) {
609 					TAILQ_REMOVE(&clp->nfsc_deleg, dp,
610 					    nfsdl_list);
611 					TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
612 					    nfsdl_list);
613 					dp->nfsdl_timestamp = NFSD_MONOSEC +
614 					    120;
615 					dp->nfsdl_rwlock.nfslock_usecnt++;
616 					*lckpp = (void *)&dp->nfsdl_rwlock;
617 				}
618 				NFSUNLOCKCLSTATE();
619 				return (0);
620 			}
621 			break;
622 		}
623 	}
624 
625 	if (p != NULL) {
626 		/*
627 		 * If p != NULL, we want to search the parentage tree
628 		 * for a matching OpenOwner and use that.
629 		 */
630 		if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
631 			nfscl_filllockowner(NULL, own, F_POSIX);
632 		else
633 			nfscl_filllockowner(p->td_proc, own, F_POSIX);
634 		nfscl_filllockowner(p->td_proc, lockown, F_POSIX);
635 		lp = NULL;
636 		error = nfscl_getopen(NULL, clp->nfsc_openhash, nfhp, fhlen,
637 		    own, lockown, mode, &lp, &op);
638 		if (error == 0 && lp != NULL && fords == 0) {
639 			/* Don't return a lock stateid for a DS. */
640 			if (NFSHASNFSV4N(nmp))
641 				stateidp->seqid = 0;
642 			else
643 				stateidp->seqid = lp->nfsl_stateid.seqid;
644 			stateidp->other[0] =
645 			    lp->nfsl_stateid.other[0];
646 			stateidp->other[1] =
647 			    lp->nfsl_stateid.other[1];
648 			stateidp->other[2] =
649 			    lp->nfsl_stateid.other[2];
650 			NFSUNLOCKCLSTATE();
651 			return (0);
652 		}
653 	}
654 	if (op == NULL) {
655 		/* If not found, just look for any OpenOwner that will work. */
656 		top = NULL;
657 		done = false;
658 		oph = NFSCLOPENHASH(clp, nfhp, fhlen);
659 		LIST_FOREACH(op, oph, nfso_hash) {
660 			if (op->nfso_fhlen == fhlen &&
661 			    !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
662 				if (top == NULL && (op->nfso_mode &
663 				    NFSV4OPEN_ACCESSWRITE) != 0 &&
664 				    (mode & NFSV4OPEN_ACCESSREAD) != 0)
665 					top = op;
666 				if ((mode & op->nfso_mode) == mode) {
667 					/* LRU order the hash list. */
668 					LIST_REMOVE(op, nfso_hash);
669 					LIST_INSERT_HEAD(oph, op, nfso_hash);
670 					done = true;
671 					break;
672 				}
673 			}
674 		}
675 		if (!done) {
676 			NFSCL_DEBUG(2, "openmode top=%p\n", top);
677 			if (top == NULL || NFSHASOPENMODE(nmp)) {
678 				NFSUNLOCKCLSTATE();
679 				return (ENOENT);
680 			} else
681 				op = top;
682 		}
683 		/*
684 		 * For read aheads or write behinds, use the open cred.
685 		 * A read ahead or write behind is indicated by p == NULL.
686 		 */
687 		if (p == NULL)
688 			newnfs_copycred(&op->nfso_cred, cred);
689 	}
690 
691 	/*
692 	 * No lock stateid, so return the open stateid.
693 	 */
694 	if (NFSHASNFSV4N(nmp))
695 		stateidp->seqid = 0;
696 	else
697 		stateidp->seqid = op->nfso_stateid.seqid;
698 	stateidp->other[0] = op->nfso_stateid.other[0];
699 	stateidp->other[1] = op->nfso_stateid.other[1];
700 	stateidp->other[2] = op->nfso_stateid.other[2];
701 	NFSUNLOCKCLSTATE();
702 	return (0);
703 }
704 
705 /*
706  * Search for a matching file, mode and, optionally, lockowner.
707  */
708 static int
709 nfscl_getopen(struct nfsclownerhead *ohp, struct nfsclopenhash *ohashp,
710     u_int8_t *nfhp, int fhlen, u_int8_t *openown, u_int8_t *lockown,
711     u_int32_t mode, struct nfscllockowner **lpp, struct nfsclopen **opp)
712 {
713 	struct nfsclowner *owp;
714 	struct nfsclopen *op, *rop, *rop2;
715 	struct nfsclopenhash *oph;
716 	bool keep_looping;
717 
718 	KASSERT(ohp == NULL || ohashp == NULL, ("nfscl_getopen: "
719 	    "only one of ohp and ohashp can be set"));
720 	if (lpp != NULL)
721 		*lpp = NULL;
722 	/*
723 	 * rop will be set to the open to be returned. There are three
724 	 * variants of this, all for an open of the correct file:
725 	 * 1 - A match of lockown.
726 	 * 2 - A match of the openown, when no lockown match exists.
727 	 * 3 - A match for any open, if no openown or lockown match exists.
728 	 * Looking for #2 over #3 probably isn't necessary, but since
729 	 * RFC3530 is vague w.r.t. the relationship between openowners and
730 	 * lockowners, I think this is the safer way to go.
731 	 */
732 	rop = NULL;
733 	rop2 = NULL;
734 	keep_looping = true;
735 	/* Search the client list */
736 	if (ohashp == NULL) {
737 		/* Search the local opens on the delegation. */
738 		LIST_FOREACH(owp, ohp, nfsow_list) {
739 			/* and look for the correct open */
740 			LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
741 				if (op->nfso_fhlen == fhlen &&
742 				    !NFSBCMP(op->nfso_fh, nfhp, fhlen)
743 				    && (op->nfso_mode & mode) == mode)
744 					keep_looping = nfscl_checkown(owp, op, openown,
745 					    lockown, lpp, &rop, &rop2);
746 				if (!keep_looping)
747 					break;
748 			}
749 			if (!keep_looping)
750 				break;
751 		}
752 	} else {
753 		/* Search for matching opens on the hash list. */
754 		oph = &ohashp[NFSCLOPENHASHFUNC(nfhp, fhlen)];
755 		LIST_FOREACH(op, oph, nfso_hash) {
756 			if (op->nfso_fhlen == fhlen &&
757 			    !NFSBCMP(op->nfso_fh, nfhp, fhlen)
758 			    && (op->nfso_mode & mode) == mode)
759 				keep_looping = nfscl_checkown(op->nfso_own, op,
760 				    openown, lockown, lpp, &rop, &rop2);
761 			if (!keep_looping) {
762 				/* LRU order the hash list. */
763 				LIST_REMOVE(op, nfso_hash);
764 				LIST_INSERT_HEAD(oph, op, nfso_hash);
765 				break;
766 			}
767 		}
768 	}
769 	if (rop == NULL)
770 		rop = rop2;
771 	if (rop == NULL)
772 		return (EBADF);
773 	*opp = rop;
774 	return (0);
775 }
776 
777 /* Check for an owner match. */
778 static bool
779 nfscl_checkown(struct nfsclowner *owp, struct nfsclopen *op, uint8_t *openown,
780     uint8_t *lockown, struct nfscllockowner **lpp, struct nfsclopen **ropp,
781     struct nfsclopen **ropp2)
782 {
783 	struct nfscllockowner *lp;
784 	bool keep_looping;
785 
786 	keep_looping = true;
787 	if (lpp != NULL) {
788 		/* Now look for a matching lockowner. */
789 		LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
790 			if (!NFSBCMP(lp->nfsl_owner, lockown,
791 			    NFSV4CL_LOCKNAMELEN)) {
792 				*lpp = lp;
793 				*ropp = op;
794 				return (false);
795 			}
796 		}
797 	}
798 	if (*ropp == NULL && !NFSBCMP(owp->nfsow_owner, openown,
799 	    NFSV4CL_LOCKNAMELEN)) {
800 		*ropp = op;
801 		if (lpp == NULL)
802 			keep_looping = false;
803 	}
804 	if (*ropp2 == NULL)
805 		*ropp2 = op;
806 	return (keep_looping);
807 }
808 
809 /*
810  * Release use of an open owner. Called when open operations are done
811  * with the open owner.
812  */
813 void
814 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
815     __unused int error, __unused int candelete, int unlocked)
816 {
817 
818 	if (owp == NULL)
819 		return;
820 	NFSLOCKCLSTATE();
821 	if (unlocked == 0) {
822 		if (NFSHASONEOPENOWN(nmp))
823 			nfsv4_relref(&owp->nfsow_rwlock);
824 		else
825 			nfscl_lockunlock(&owp->nfsow_rwlock);
826 	}
827 	nfscl_clrelease(owp->nfsow_clp);
828 	NFSUNLOCKCLSTATE();
829 }
830 
831 /*
832  * Release use of an open structure under an open owner.
833  */
834 void
835 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
836     int candelete)
837 {
838 	struct nfsclclient *clp;
839 	struct nfsclowner *owp;
840 
841 	if (op == NULL)
842 		return;
843 	NFSLOCKCLSTATE();
844 	owp = op->nfso_own;
845 	if (NFSHASONEOPENOWN(nmp))
846 		nfsv4_relref(&owp->nfsow_rwlock);
847 	else
848 		nfscl_lockunlock(&owp->nfsow_rwlock);
849 	clp = owp->nfsow_clp;
850 	if (error && candelete && op->nfso_opencnt == 0)
851 		nfscl_freeopen(op, 0, true);
852 	nfscl_clrelease(clp);
853 	NFSUNLOCKCLSTATE();
854 }
855 
856 /*
857  * Called to get a clientid structure. It will optionally lock the
858  * client data structures to do the SetClientId/SetClientId_confirm,
859  * but will release that lock and return the clientid with a reference
860  * count on it.
861  * If the "cred" argument is NULL, a new clientid should not be created.
862  * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
863  * be done.
864  * It always clpp with a reference count on it, unless returning an error.
865  */
866 int
867 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
868     bool tryminvers, bool firstref, struct nfsclclient **clpp)
869 {
870 	struct nfsclclient *clp;
871 	struct nfsclclient *newclp = NULL;
872 	struct nfsmount *nmp;
873 	char uuid[HOSTUUIDLEN];
874 	int igotlock = 0, error, trystalecnt, clidinusedelay, i;
875 	u_int16_t idlen = 0;
876 
877 	nmp = VFSTONFS(mp);
878 	if (cred != NULL) {
879 		getcredhostuuid(cred, uuid, sizeof uuid);
880 		idlen = strlen(uuid);
881 		if (idlen > 0)
882 			idlen += sizeof (u_int64_t);
883 		else
884 			idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
885 		newclp = malloc(
886 		    sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
887 		    M_WAITOK | M_ZERO);
888 	}
889 	NFSLOCKCLSTATE();
890 	/*
891 	 * If a forced dismount is already in progress, don't
892 	 * allocate a new clientid and get out now. For the case where
893 	 * clp != NULL, this is a harmless optimization.
894 	 */
895 	if (NFSCL_FORCEDISM(mp)) {
896 		NFSUNLOCKCLSTATE();
897 		if (newclp != NULL)
898 			free(newclp, M_NFSCLCLIENT);
899 		return (EBADF);
900 	}
901 	clp = nmp->nm_clp;
902 	if (clp == NULL) {
903 		if (newclp == NULL) {
904 			NFSUNLOCKCLSTATE();
905 			return (EACCES);
906 		}
907 		clp = newclp;
908 		clp->nfsc_idlen = idlen;
909 		LIST_INIT(&clp->nfsc_owner);
910 		TAILQ_INIT(&clp->nfsc_deleg);
911 		TAILQ_INIT(&clp->nfsc_layout);
912 		LIST_INIT(&clp->nfsc_devinfo);
913 		for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
914 			LIST_INIT(&clp->nfsc_deleghash[i]);
915 		for (i = 0; i < NFSCLOPENHASHSIZE; i++)
916 			LIST_INIT(&clp->nfsc_openhash[i]);
917 		for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
918 			LIST_INIT(&clp->nfsc_layouthash[i]);
919 		clp->nfsc_flags = NFSCLFLAGS_INITED;
920 		clp->nfsc_clientidrev = 1;
921 		clp->nfsc_cbident = nfscl_nextcbident();
922 		nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
923 		    clp->nfsc_idlen);
924 		LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
925 		nmp->nm_clp = clp;
926 		clp->nfsc_nmp = nmp;
927 	} else {
928 		if (newclp != NULL)
929 			free(newclp, M_NFSCLCLIENT);
930 	}
931 	while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
932 	    !NFSCL_FORCEDISM(mp))
933 		igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
934 		    NFSCLSTATEMUTEXPTR, mp);
935 	if (igotlock == 0) {
936 		/*
937 		 * Call nfsv4_lock() with "iwantlock == 0" on the firstref so
938 		 * that it will wait for a pending exclusive lock request.
939 		 * This gives the exclusive lock request priority over this
940 		 * shared lock request.
941 		 * An exclusive lock on nfsc_lock is used mainly for server
942 		 * crash recoveries and delegation recalls.
943 		 */
944 		if (firstref)
945 			nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR,
946 			    mp);
947 		nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
948 	}
949 	if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
950 		/*
951 		 * Both nfsv4_lock() and nfsv4_getref() know to check
952 		 * for NFSCL_FORCEDISM() and return without sleeping to
953 		 * wait for the exclusive lock to be released, since it
954 		 * might be held by nfscl_umount() and we need to get out
955 		 * now for that case and not wait until nfscl_umount()
956 		 * releases it.
957 		 */
958 		NFSUNLOCKCLSTATE();
959 		return (EBADF);
960 	}
961 	NFSUNLOCKCLSTATE();
962 
963 	/*
964 	 * If it needs a clientid, do the setclientid now.
965 	 */
966 	if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
967 		if (!igotlock)
968 			panic("nfscl_clget");
969 		if (p == NULL || cred == NULL) {
970 			NFSLOCKCLSTATE();
971 			nfsv4_unlock(&clp->nfsc_lock, 0);
972 			NFSUNLOCKCLSTATE();
973 			return (EACCES);
974 		}
975 		/*
976 		 * If RFC3530 Sec. 14.2.33 is taken literally,
977 		 * NFSERR_CLIDINUSE will be returned persistently for the
978 		 * case where a new mount of the same file system is using
979 		 * a different principal. In practice, NFSERR_CLIDINUSE is
980 		 * only returned when there is outstanding unexpired state
981 		 * on the clientid. As such, try for twice the lease
982 		 * interval, if we know what that is. Otherwise, make a
983 		 * wild ass guess.
984 		 * The case of returning NFSERR_STALECLIENTID is far less
985 		 * likely, but might occur if there is a significant delay
986 		 * between doing the SetClientID and SetClientIDConfirm Ops,
987 		 * such that the server throws away the clientid before
988 		 * receiving the SetClientIDConfirm.
989 		 */
990 		if (clp->nfsc_renew > 0)
991 			clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
992 		else
993 			clidinusedelay = 120;
994 		trystalecnt = 3;
995 		do {
996 			error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
997 			if (error == NFSERR_STALECLIENTID ||
998 			    error == NFSERR_STALEDONTRECOVER ||
999 			    error == NFSERR_BADSESSION ||
1000 			    error == NFSERR_CLIDINUSE) {
1001 				(void) nfs_catnap(PZERO, error, "nfs_setcl");
1002 			} else if (error == NFSERR_MINORVERMISMATCH &&
1003 			    tryminvers) {
1004 				if (nmp->nm_minorvers > 0)
1005 					nmp->nm_minorvers--;
1006 				else
1007 					tryminvers = false;
1008 			}
1009 		} while (((error == NFSERR_STALECLIENTID ||
1010 		     error == NFSERR_BADSESSION ||
1011 		     error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
1012 		    (error == NFSERR_CLIDINUSE && --clidinusedelay > 0) ||
1013 		    (error == NFSERR_MINORVERMISMATCH && tryminvers));
1014 		if (error) {
1015 			NFSLOCKCLSTATE();
1016 			nfsv4_unlock(&clp->nfsc_lock, 0);
1017 			NFSUNLOCKCLSTATE();
1018 			return (error);
1019 		}
1020 		clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
1021 	}
1022 	if (igotlock) {
1023 		NFSLOCKCLSTATE();
1024 		nfsv4_unlock(&clp->nfsc_lock, 1);
1025 		NFSUNLOCKCLSTATE();
1026 	}
1027 
1028 	*clpp = clp;
1029 	return (0);
1030 }
1031 
1032 /*
1033  * Get a reference to a clientid and return it, if valid.
1034  */
1035 struct nfsclclient *
1036 nfscl_findcl(struct nfsmount *nmp)
1037 {
1038 	struct nfsclclient *clp;
1039 
1040 	clp = nmp->nm_clp;
1041 	if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
1042 		return (NULL);
1043 	return (clp);
1044 }
1045 
1046 /*
1047  * Release the clientid structure. It may be locked or reference counted.
1048  */
1049 static void
1050 nfscl_clrelease(struct nfsclclient *clp)
1051 {
1052 
1053 	if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
1054 		nfsv4_unlock(&clp->nfsc_lock, 0);
1055 	else
1056 		nfsv4_relref(&clp->nfsc_lock);
1057 }
1058 
1059 /*
1060  * External call for nfscl_clrelease.
1061  */
1062 void
1063 nfscl_clientrelease(struct nfsclclient *clp)
1064 {
1065 
1066 	NFSLOCKCLSTATE();
1067 	if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
1068 		nfsv4_unlock(&clp->nfsc_lock, 0);
1069 	else
1070 		nfsv4_relref(&clp->nfsc_lock);
1071 	NFSUNLOCKCLSTATE();
1072 }
1073 
1074 /*
1075  * Called when wanting to lock a byte region.
1076  */
1077 int
1078 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1079     short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
1080     int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
1081     struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
1082 {
1083 	struct nfscllockowner *lp;
1084 	struct nfsclopen *op;
1085 	struct nfsclclient *clp;
1086 	struct nfscllockowner *nlp;
1087 	struct nfscllock *nlop, *otherlop;
1088 	struct nfscldeleg *dp = NULL, *ldp = NULL;
1089 	struct nfscllockownerhead *lhp = NULL;
1090 	struct nfsnode *np;
1091 	u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
1092 	u_int8_t *openownp;
1093 	int error = 0, ret, donelocally = 0;
1094 	u_int32_t mode;
1095 
1096 	/* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1097 	mode = 0;
1098 	np = VTONFS(vp);
1099 	*lpp = NULL;
1100 	lp = NULL;
1101 	*newonep = 0;
1102 	*donelocallyp = 0;
1103 
1104 	/*
1105 	 * Might need these, so MALLOC them now, to
1106 	 * avoid a tsleep() in MALLOC later.
1107 	 */
1108 	nlp = malloc(
1109 	    sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1110 	otherlop = malloc(
1111 	    sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1112 	nlop = malloc(
1113 	    sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1114 	nlop->nfslo_type = type;
1115 	nlop->nfslo_first = off;
1116 	if (len == NFS64BITSSET) {
1117 		nlop->nfslo_end = NFS64BITSSET;
1118 	} else {
1119 		nlop->nfslo_end = off + len;
1120 		if (nlop->nfslo_end <= nlop->nfslo_first)
1121 			error = NFSERR_INVAL;
1122 	}
1123 
1124 	if (!error) {
1125 		if (recovery)
1126 			clp = rclp;
1127 		else
1128 			error = nfscl_getcl(vp->v_mount, cred, p, false, true,
1129 			    &clp);
1130 	}
1131 	if (error) {
1132 		free(nlp, M_NFSCLLOCKOWNER);
1133 		free(otherlop, M_NFSCLLOCK);
1134 		free(nlop, M_NFSCLLOCK);
1135 		return (error);
1136 	}
1137 
1138 	op = NULL;
1139 	if (recovery) {
1140 		ownp = rownp;
1141 		openownp = ropenownp;
1142 	} else {
1143 		nfscl_filllockowner(id, own, flags);
1144 		ownp = own;
1145 		if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
1146 			nfscl_filllockowner(NULL, openown, F_POSIX);
1147 		else
1148 			nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1149 		openownp = openown;
1150 	}
1151 	if (!recovery) {
1152 		NFSLOCKCLSTATE();
1153 		/*
1154 		 * First, search for a delegation. If one exists for this file,
1155 		 * the lock can be done locally against it, so long as there
1156 		 * isn't a local lock conflict.
1157 		 */
1158 		ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1159 		    np->n_fhp->nfh_len);
1160 		/* Just sanity check for correct type of delegation */
1161 		if (dp != NULL && ((dp->nfsdl_flags &
1162 		    (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1163 		     (type == F_WRLCK &&
1164 		      (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1165 			dp = NULL;
1166 	}
1167 	if (dp != NULL) {
1168 		/* Now, find an open and maybe a lockowner. */
1169 		ret = nfscl_getopen(&dp->nfsdl_owner, NULL, np->n_fhp->nfh_fh,
1170 		    np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1171 		if (ret)
1172 			ret = nfscl_getopen(NULL, clp->nfsc_openhash,
1173 			    np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1174 			    ownp, mode, NULL, &op);
1175 		if (!ret) {
1176 			lhp = &dp->nfsdl_lock;
1177 			TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1178 			TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1179 			dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1180 			donelocally = 1;
1181 		} else {
1182 			dp = NULL;
1183 		}
1184 	}
1185 	if (!donelocally) {
1186 		/*
1187 		 * Get the related Open and maybe lockowner.
1188 		 */
1189 		error = nfscl_getopen(NULL, clp->nfsc_openhash,
1190 		    np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1191 		    ownp, mode, &lp, &op);
1192 		if (!error)
1193 			lhp = &op->nfso_lock;
1194 	}
1195 	if (!error && !recovery)
1196 		error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1197 		    np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1198 	if (error) {
1199 		if (!recovery) {
1200 			nfscl_clrelease(clp);
1201 			NFSUNLOCKCLSTATE();
1202 		}
1203 		free(nlp, M_NFSCLLOCKOWNER);
1204 		free(otherlop, M_NFSCLLOCK);
1205 		free(nlop, M_NFSCLLOCK);
1206 		return (error);
1207 	}
1208 
1209 	/*
1210 	 * Ok, see if a lockowner exists and create one, as required.
1211 	 */
1212 	if (lp == NULL)
1213 		LIST_FOREACH(lp, lhp, nfsl_list) {
1214 			if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1215 				break;
1216 		}
1217 	if (lp == NULL) {
1218 		NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1219 		if (recovery)
1220 			NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1221 			    NFSV4CL_LOCKNAMELEN);
1222 		else
1223 			NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1224 			    NFSV4CL_LOCKNAMELEN);
1225 		nlp->nfsl_seqid = 0;
1226 		nlp->nfsl_lockflags = flags;
1227 		nlp->nfsl_inprog = NULL;
1228 		nfscl_lockinit(&nlp->nfsl_rwlock);
1229 		LIST_INIT(&nlp->nfsl_lock);
1230 		if (donelocally) {
1231 			nlp->nfsl_open = NULL;
1232 			nfsstatsv1.cllocallockowners++;
1233 		} else {
1234 			nlp->nfsl_open = op;
1235 			nfsstatsv1.cllockowners++;
1236 		}
1237 		LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1238 		lp = nlp;
1239 		nlp = NULL;
1240 		*newonep = 1;
1241 	}
1242 
1243 	/*
1244 	 * Now, update the byte ranges for locks.
1245 	 */
1246 	ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1247 	if (!ret)
1248 		donelocally = 1;
1249 	if (donelocally) {
1250 		*donelocallyp = 1;
1251 		if (!recovery)
1252 			nfscl_clrelease(clp);
1253 	} else {
1254 		/*
1255 		 * Serial modifications on the lock owner for multiple threads
1256 		 * for the same process using a read/write lock.
1257 		 */
1258 		if (!recovery)
1259 			nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1260 	}
1261 	if (!recovery)
1262 		NFSUNLOCKCLSTATE();
1263 
1264 	if (nlp)
1265 		free(nlp, M_NFSCLLOCKOWNER);
1266 	if (nlop)
1267 		free(nlop, M_NFSCLLOCK);
1268 	if (otherlop)
1269 		free(otherlop, M_NFSCLLOCK);
1270 
1271 	*lpp = lp;
1272 	return (0);
1273 }
1274 
1275 /*
1276  * Called to unlock a byte range, for LockU.
1277  */
1278 int
1279 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1280     __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1281     struct nfsclclient *clp, void *id, int flags,
1282     struct nfscllockowner **lpp, int *dorpcp)
1283 {
1284 	struct nfscllockowner *lp;
1285 	struct nfsclopen *op;
1286 	struct nfscllock *nlop, *other_lop = NULL;
1287 	struct nfscldeleg *dp;
1288 	struct nfsnode *np;
1289 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
1290 	int ret = 0, fnd;
1291 
1292 	np = VTONFS(vp);
1293 	*lpp = NULL;
1294 	*dorpcp = 0;
1295 
1296 	/*
1297 	 * Might need these, so MALLOC them now, to
1298 	 * avoid a tsleep() in MALLOC later.
1299 	 */
1300 	nlop = malloc(
1301 	    sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1302 	nlop->nfslo_type = F_UNLCK;
1303 	nlop->nfslo_first = off;
1304 	if (len == NFS64BITSSET) {
1305 		nlop->nfslo_end = NFS64BITSSET;
1306 	} else {
1307 		nlop->nfslo_end = off + len;
1308 		if (nlop->nfslo_end <= nlop->nfslo_first) {
1309 			free(nlop, M_NFSCLLOCK);
1310 			return (NFSERR_INVAL);
1311 		}
1312 	}
1313 	if (callcnt == 0) {
1314 		other_lop = malloc(
1315 		    sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1316 		*other_lop = *nlop;
1317 	}
1318 	nfscl_filllockowner(id, own, flags);
1319 	dp = NULL;
1320 	NFSLOCKCLSTATE();
1321 	if (callcnt == 0)
1322 		dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1323 		    np->n_fhp->nfh_len);
1324 
1325 	/*
1326 	 * First, unlock any local regions on a delegation.
1327 	 */
1328 	if (dp != NULL) {
1329 		/* Look for this lockowner. */
1330 		LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1331 			if (!NFSBCMP(lp->nfsl_owner, own,
1332 			    NFSV4CL_LOCKNAMELEN))
1333 				break;
1334 		}
1335 		if (lp != NULL)
1336 			/* Use other_lop, so nlop is still available */
1337 			(void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1338 	}
1339 
1340 	/*
1341 	 * Now, find a matching open/lockowner that hasn't already been done,
1342 	 * as marked by nfsl_inprog.
1343 	 */
1344 	lp = NULL;
1345 	fnd = 0;
1346 	LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1347 	    np->n_fhp->nfh_len), nfso_hash) {
1348 		if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1349 		    !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1350 			LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1351 				if (lp->nfsl_inprog == NULL &&
1352 				    !NFSBCMP(lp->nfsl_owner, own,
1353 				     NFSV4CL_LOCKNAMELEN)) {
1354 					fnd = 1;
1355 					break;
1356 				}
1357 			}
1358 		}
1359 		if (fnd)
1360 			break;
1361 	}
1362 
1363 	if (lp != NULL) {
1364 		ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1365 		if (ret)
1366 			*dorpcp = 1;
1367 		/*
1368 		 * Serial modifications on the lock owner for multiple
1369 		 * threads for the same process using a read/write lock.
1370 		 */
1371 		lp->nfsl_inprog = p;
1372 		nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1373 		*lpp = lp;
1374 	}
1375 	NFSUNLOCKCLSTATE();
1376 	if (nlop)
1377 		free(nlop, M_NFSCLLOCK);
1378 	if (other_lop)
1379 		free(other_lop, M_NFSCLLOCK);
1380 	return (0);
1381 }
1382 
1383 /*
1384  * Release all lockowners marked in progess for this process and file.
1385  */
1386 void
1387 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1388     void *id, int flags)
1389 {
1390 	struct nfsclopen *op;
1391 	struct nfscllockowner *lp;
1392 	struct nfsnode *np;
1393 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
1394 
1395 	np = VTONFS(vp);
1396 	nfscl_filllockowner(id, own, flags);
1397 	NFSLOCKCLSTATE();
1398 	LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1399 	    np->n_fhp->nfh_len), nfso_hash) {
1400 		if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1401 		    !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1402 			LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1403 				if (lp->nfsl_inprog == p &&
1404 				    !NFSBCMP(lp->nfsl_owner, own,
1405 				    NFSV4CL_LOCKNAMELEN)) {
1406 					lp->nfsl_inprog = NULL;
1407 					nfscl_lockunlock(&lp->nfsl_rwlock);
1408 				}
1409 			}
1410 		}
1411 	}
1412 	nfscl_clrelease(clp);
1413 	NFSUNLOCKCLSTATE();
1414 }
1415 
1416 /*
1417  * Called to find out if any bytes within the byte range specified are
1418  * write locked by the calling process. Used to determine if flushing
1419  * is required before a LockU.
1420  * If in doubt, return 1, so the flush will occur.
1421  */
1422 int
1423 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1424     struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1425 {
1426 	struct nfscllockowner *lp;
1427 	struct nfsclopen *op;
1428 	struct nfsclclient *clp;
1429 	struct nfscllock *lop;
1430 	struct nfscldeleg *dp;
1431 	struct nfsnode *np;
1432 	u_int64_t off, end;
1433 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
1434 	int error = 0;
1435 
1436 	np = VTONFS(vp);
1437 	switch (fl->l_whence) {
1438 	case SEEK_SET:
1439 	case SEEK_CUR:
1440 		/*
1441 		 * Caller is responsible for adding any necessary offset
1442 		 * when SEEK_CUR is used.
1443 		 */
1444 		off = fl->l_start;
1445 		break;
1446 	case SEEK_END:
1447 		off = np->n_size + fl->l_start;
1448 		break;
1449 	default:
1450 		return (1);
1451 	}
1452 	if (fl->l_len != 0) {
1453 		end = off + fl->l_len;
1454 		if (end < off)
1455 			return (1);
1456 	} else {
1457 		end = NFS64BITSSET;
1458 	}
1459 
1460 	error = nfscl_getcl(vp->v_mount, cred, p, false, true, &clp);
1461 	if (error)
1462 		return (1);
1463 	nfscl_filllockowner(id, own, flags);
1464 	NFSLOCKCLSTATE();
1465 
1466 	/*
1467 	 * First check the delegation locks.
1468 	 */
1469 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1470 	if (dp != NULL) {
1471 		LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1472 			if (!NFSBCMP(lp->nfsl_owner, own,
1473 			    NFSV4CL_LOCKNAMELEN))
1474 				break;
1475 		}
1476 		if (lp != NULL) {
1477 			LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1478 				if (lop->nfslo_first >= end)
1479 					break;
1480 				if (lop->nfslo_end <= off)
1481 					continue;
1482 				if (lop->nfslo_type == F_WRLCK) {
1483 					nfscl_clrelease(clp);
1484 					NFSUNLOCKCLSTATE();
1485 					return (1);
1486 				}
1487 			}
1488 		}
1489 	}
1490 
1491 	/*
1492 	 * Now, check state against the server.
1493 	 */
1494 	LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1495 	    np->n_fhp->nfh_len), nfso_hash) {
1496 		if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1497 		    !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1498 			LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1499 				if (!NFSBCMP(lp->nfsl_owner, own,
1500 				    NFSV4CL_LOCKNAMELEN))
1501 					break;
1502 			}
1503 			if (lp != NULL) {
1504 				LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1505 					if (lop->nfslo_first >= end)
1506 						break;
1507 					if (lop->nfslo_end <= off)
1508 						continue;
1509 					if (lop->nfslo_type == F_WRLCK) {
1510 						nfscl_clrelease(clp);
1511 						NFSUNLOCKCLSTATE();
1512 						return (1);
1513 					}
1514 				}
1515 			}
1516 		}
1517 	}
1518 	nfscl_clrelease(clp);
1519 	NFSUNLOCKCLSTATE();
1520 	return (0);
1521 }
1522 
1523 /*
1524  * Release a byte range lock owner structure.
1525  */
1526 void
1527 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1528 {
1529 	struct nfsclclient *clp;
1530 
1531 	if (lp == NULL)
1532 		return;
1533 	NFSLOCKCLSTATE();
1534 	clp = lp->nfsl_open->nfso_own->nfsow_clp;
1535 	if (error != 0 && candelete &&
1536 	    (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1537 		nfscl_freelockowner(lp, 0);
1538 	else
1539 		nfscl_lockunlock(&lp->nfsl_rwlock);
1540 	nfscl_clrelease(clp);
1541 	NFSUNLOCKCLSTATE();
1542 }
1543 
1544 /*
1545  * Unlink the open structure.
1546  */
1547 static void
1548 nfscl_unlinkopen(struct nfsclopen *op)
1549 {
1550 
1551 	LIST_REMOVE(op, nfso_list);
1552 	if (op->nfso_hash.le_prev != NULL)
1553 		LIST_REMOVE(op, nfso_hash);
1554 }
1555 
1556 /*
1557  * Free up an open structure and any associated byte range lock structures.
1558  */
1559 void
1560 nfscl_freeopen(struct nfsclopen *op, int local, bool unlink)
1561 {
1562 
1563 	if (unlink)
1564 		nfscl_unlinkopen(op);
1565 	nfscl_freealllocks(&op->nfso_lock, local);
1566 	free(op, M_NFSCLOPEN);
1567 	if (local)
1568 		nfsstatsv1.cllocalopens--;
1569 	else
1570 		nfsstatsv1.clopens--;
1571 }
1572 
1573 /*
1574  * Free up all lock owners and associated locks.
1575  */
1576 static void
1577 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1578 {
1579 	struct nfscllockowner *lp, *nlp;
1580 
1581 	LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1582 		if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1583 			panic("nfscllckw");
1584 		nfscl_freelockowner(lp, local);
1585 	}
1586 }
1587 
1588 /*
1589  * Called for an Open when NFSERR_EXPIRED is received from the server.
1590  * If there are no byte range locks nor a Share Deny lost, try to do a
1591  * fresh Open. Otherwise, free the open.
1592  */
1593 static int
1594 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1595     struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1596 {
1597 	struct nfscllockowner *lp;
1598 	struct nfscldeleg *dp;
1599 	int mustdelete = 0, error;
1600 
1601 	/*
1602 	 * Look for any byte range lock(s).
1603 	 */
1604 	LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1605 		if (!LIST_EMPTY(&lp->nfsl_lock)) {
1606 			mustdelete = 1;
1607 			break;
1608 		}
1609 	}
1610 
1611 	/*
1612 	 * If no byte range lock(s) nor a Share deny, try to re-open.
1613 	 */
1614 	if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1615 		newnfs_copycred(&op->nfso_cred, cred);
1616 		dp = NULL;
1617 		error = nfsrpc_reopen(nmp, op->nfso_fh,
1618 		    op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1619 		if (error) {
1620 			mustdelete = 1;
1621 			if (dp != NULL) {
1622 				free(dp, M_NFSCLDELEG);
1623 				dp = NULL;
1624 			}
1625 		}
1626 		if (dp != NULL)
1627 			nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1628 			    op->nfso_fhlen, cred, p, &dp);
1629 	}
1630 
1631 	/*
1632 	 * If a byte range lock or Share deny or couldn't re-open, free it.
1633 	 */
1634 	if (mustdelete)
1635 		nfscl_freeopen(op, 0, true);
1636 	return (mustdelete);
1637 }
1638 
1639 /*
1640  * Free up an open owner structure.
1641  */
1642 static void
1643 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1644 {
1645 
1646 	LIST_REMOVE(owp, nfsow_list);
1647 	free(owp, M_NFSCLOWNER);
1648 	if (local)
1649 		nfsstatsv1.cllocalopenowners--;
1650 	else
1651 		nfsstatsv1.clopenowners--;
1652 }
1653 
1654 /*
1655  * Free up a byte range lock owner structure.
1656  */
1657 void
1658 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1659 {
1660 	struct nfscllock *lop, *nlop;
1661 
1662 	LIST_REMOVE(lp, nfsl_list);
1663 	LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1664 		nfscl_freelock(lop, local);
1665 	}
1666 	free(lp, M_NFSCLLOCKOWNER);
1667 	if (local)
1668 		nfsstatsv1.cllocallockowners--;
1669 	else
1670 		nfsstatsv1.cllockowners--;
1671 }
1672 
1673 /*
1674  * Free up a byte range lock structure.
1675  */
1676 void
1677 nfscl_freelock(struct nfscllock *lop, int local)
1678 {
1679 
1680 	LIST_REMOVE(lop, nfslo_list);
1681 	free(lop, M_NFSCLLOCK);
1682 	if (local)
1683 		nfsstatsv1.cllocallocks--;
1684 	else
1685 		nfsstatsv1.cllocks--;
1686 }
1687 
1688 /*
1689  * Clean out the state related to a delegation.
1690  */
1691 static void
1692 nfscl_cleandeleg(struct nfscldeleg *dp)
1693 {
1694 	struct nfsclowner *owp, *nowp;
1695 	struct nfsclopen *op;
1696 
1697 	LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1698 		op = LIST_FIRST(&owp->nfsow_open);
1699 		if (op != NULL) {
1700 			if (LIST_NEXT(op, nfso_list) != NULL)
1701 				panic("nfscleandel");
1702 			nfscl_freeopen(op, 1, true);
1703 		}
1704 		nfscl_freeopenowner(owp, 1);
1705 	}
1706 	nfscl_freealllocks(&dp->nfsdl_lock, 1);
1707 }
1708 
1709 /*
1710  * Free a delegation.
1711  */
1712 static void
1713 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp, bool freeit)
1714 {
1715 
1716 	TAILQ_REMOVE(hdp, dp, nfsdl_list);
1717 	LIST_REMOVE(dp, nfsdl_hash);
1718 	if (freeit)
1719 		free(dp, M_NFSCLDELEG);
1720 	nfsstatsv1.cldelegates--;
1721 	nfscl_delegcnt--;
1722 }
1723 
1724 /*
1725  * Free up all state related to this client structure.
1726  */
1727 static void
1728 nfscl_cleanclient(struct nfsclclient *clp)
1729 {
1730 	struct nfsclowner *owp, *nowp;
1731 	struct nfsclopen *op, *nop;
1732 	struct nfscllayout *lyp, *nlyp;
1733 	struct nfscldevinfo *dip, *ndip;
1734 
1735 	TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1736 		nfscl_freelayout(lyp);
1737 
1738 	LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1739 		nfscl_freedevinfo(dip);
1740 
1741 	/* Now, all the OpenOwners, etc. */
1742 	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1743 		LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1744 			nfscl_freeopen(op, 0, true);
1745 		}
1746 		nfscl_freeopenowner(owp, 0);
1747 	}
1748 }
1749 
1750 /*
1751  * Called when an NFSERR_EXPIRED is received from the server.
1752  */
1753 static void
1754 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1755     struct ucred *cred, NFSPROC_T *p)
1756 {
1757 	struct nfsclowner *owp, *nowp, *towp;
1758 	struct nfsclopen *op, *nop, *top;
1759 	struct nfscldeleg *dp, *ndp;
1760 	int ret, printed = 0;
1761 
1762 	/*
1763 	 * First, merge locally issued Opens into the list for the server.
1764 	 */
1765 	dp = TAILQ_FIRST(&clp->nfsc_deleg);
1766 	while (dp != NULL) {
1767 	    ndp = TAILQ_NEXT(dp, nfsdl_list);
1768 	    owp = LIST_FIRST(&dp->nfsdl_owner);
1769 	    while (owp != NULL) {
1770 		nowp = LIST_NEXT(owp, nfsow_list);
1771 		op = LIST_FIRST(&owp->nfsow_open);
1772 		if (op != NULL) {
1773 		    if (LIST_NEXT(op, nfso_list) != NULL)
1774 			panic("nfsclexp");
1775 		    LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1776 			if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1777 			    NFSV4CL_LOCKNAMELEN))
1778 			    break;
1779 		    }
1780 		    if (towp != NULL) {
1781 			/* Merge opens in */
1782 			LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1783 			    if (top->nfso_fhlen == op->nfso_fhlen &&
1784 				!NFSBCMP(top->nfso_fh, op->nfso_fh,
1785 				 op->nfso_fhlen)) {
1786 				top->nfso_mode |= op->nfso_mode;
1787 				top->nfso_opencnt += op->nfso_opencnt;
1788 				break;
1789 			    }
1790 			}
1791 			if (top == NULL) {
1792 			    /* Just add the open to the owner list */
1793 			    LIST_REMOVE(op, nfso_list);
1794 			    op->nfso_own = towp;
1795 			    LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1796 			    LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
1797 				op->nfso_fhlen), op, nfso_hash);
1798 			    nfsstatsv1.cllocalopens--;
1799 			    nfsstatsv1.clopens++;
1800 			}
1801 		    } else {
1802 			/* Just add the openowner to the client list */
1803 			LIST_REMOVE(owp, nfsow_list);
1804 			owp->nfsow_clp = clp;
1805 			LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1806 			LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
1807 			    op->nfso_fhlen), op, nfso_hash);
1808 			nfsstatsv1.cllocalopenowners--;
1809 			nfsstatsv1.clopenowners++;
1810 			nfsstatsv1.cllocalopens--;
1811 			nfsstatsv1.clopens++;
1812 		    }
1813 		}
1814 		owp = nowp;
1815 	    }
1816 	    if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1817 		printed = 1;
1818 		printf("nfsv4 expired locks lost\n");
1819 	    }
1820 	    nfscl_cleandeleg(dp);
1821 	    nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
1822 	    dp = ndp;
1823 	}
1824 	if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1825 	    panic("nfsclexp");
1826 
1827 	/*
1828 	 * Now, try and reopen against the server.
1829 	 */
1830 	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1831 		owp->nfsow_seqid = 0;
1832 		LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1833 			ret = nfscl_expireopen(clp, op, nmp, cred, p);
1834 			if (ret && !printed) {
1835 				printed = 1;
1836 				printf("nfsv4 expired locks lost\n");
1837 			}
1838 		}
1839 		if (LIST_EMPTY(&owp->nfsow_open))
1840 			nfscl_freeopenowner(owp, 0);
1841 	}
1842 }
1843 
1844 /*
1845  * This function must be called after the process represented by "own" has
1846  * exited. Must be called with CLSTATE lock held.
1847  */
1848 static void
1849 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1850 {
1851 	struct nfsclowner *owp, *nowp;
1852 	struct nfscllockowner *lp, *nlp;
1853 	struct nfscldeleg *dp;
1854 
1855 	/* First, get rid of local locks on delegations. */
1856 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1857 		LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1858 		    if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1859 			if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1860 			    panic("nfscllckw");
1861 			nfscl_freelockowner(lp, 1);
1862 		    }
1863 		}
1864 	}
1865 	owp = LIST_FIRST(&clp->nfsc_owner);
1866 	while (owp != NULL) {
1867 		nowp = LIST_NEXT(owp, nfsow_list);
1868 		if (!NFSBCMP(owp->nfsow_owner, own,
1869 		    NFSV4CL_LOCKNAMELEN)) {
1870 			/*
1871 			 * If there are children that haven't closed the
1872 			 * file descriptors yet, the opens will still be
1873 			 * here. For that case, let the renew thread clear
1874 			 * out the OpenOwner later.
1875 			 */
1876 			if (LIST_EMPTY(&owp->nfsow_open))
1877 				nfscl_freeopenowner(owp, 0);
1878 			else
1879 				owp->nfsow_defunct = 1;
1880 		}
1881 		owp = nowp;
1882 	}
1883 }
1884 
1885 /*
1886  * Find open/lock owners for processes that have exited.
1887  */
1888 static void
1889 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1890 {
1891 	struct nfsclowner *owp, *nowp;
1892 	struct nfsclopen *op;
1893 	struct nfscllockowner *lp, *nlp;
1894 	struct nfscldeleg *dp;
1895 
1896 	/*
1897 	 * All the pidhash locks must be acquired, since they are sx locks
1898 	 * and must be acquired before the mutexes.  The pid(s) that will
1899 	 * be used aren't known yet, so all the locks need to be acquired.
1900 	 * Fortunately, this function is only performed once/sec.
1901 	 */
1902 	pidhash_slockall();
1903 	NFSLOCKCLSTATE();
1904 	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1905 		LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1906 			LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1907 				if (LIST_EMPTY(&lp->nfsl_lock))
1908 					nfscl_emptylockowner(lp, lhp);
1909 			}
1910 		}
1911 		if (nfscl_procdoesntexist(owp->nfsow_owner))
1912 			nfscl_cleanup_common(clp, owp->nfsow_owner);
1913 	}
1914 
1915 	/*
1916 	 * For the single open_owner case, these lock owners need to be
1917 	 * checked to see if they still exist separately.
1918 	 * This is because nfscl_procdoesntexist() never returns true for
1919 	 * the single open_owner so that the above doesn't ever call
1920 	 * nfscl_cleanup_common().
1921 	 */
1922 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1923 		LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1924 			if (nfscl_procdoesntexist(lp->nfsl_owner))
1925 				nfscl_cleanup_common(clp, lp->nfsl_owner);
1926 		}
1927 	}
1928 	NFSUNLOCKCLSTATE();
1929 	pidhash_sunlockall();
1930 }
1931 
1932 /*
1933  * Take the empty lock owner and move it to the local lhp list if the
1934  * associated process no longer exists.
1935  */
1936 static void
1937 nfscl_emptylockowner(struct nfscllockowner *lp,
1938     struct nfscllockownerfhhead *lhp)
1939 {
1940 	struct nfscllockownerfh *lfhp, *mylfhp;
1941 	struct nfscllockowner *nlp;
1942 	int fnd_it;
1943 
1944 	/* If not a Posix lock owner, just return. */
1945 	if ((lp->nfsl_lockflags & F_POSIX) == 0)
1946 		return;
1947 
1948 	fnd_it = 0;
1949 	mylfhp = NULL;
1950 	/*
1951 	 * First, search to see if this lock owner is already in the list.
1952 	 * If it is, then the associated process no longer exists.
1953 	 */
1954 	SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1955 		if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1956 		    !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1957 		    lfhp->nfslfh_len))
1958 			mylfhp = lfhp;
1959 		LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1960 			if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1961 			    NFSV4CL_LOCKNAMELEN))
1962 				fnd_it = 1;
1963 	}
1964 	/* If not found, check if process still exists. */
1965 	if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1966 		return;
1967 
1968 	/* Move the lock owner over to the local list. */
1969 	if (mylfhp == NULL) {
1970 		mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1971 		    M_NOWAIT);
1972 		if (mylfhp == NULL)
1973 			return;
1974 		mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1975 		NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1976 		    mylfhp->nfslfh_len);
1977 		LIST_INIT(&mylfhp->nfslfh_lock);
1978 		SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1979 	}
1980 	LIST_REMOVE(lp, nfsl_list);
1981 	LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1982 }
1983 
1984 static int	fake_global;	/* Used to force visibility of MNTK_UNMOUNTF */
1985 /*
1986  * Called from nfs umount to free up the clientid.
1987  */
1988 void
1989 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p, struct nfscldeleghead *dhp)
1990 {
1991 	struct nfsclclient *clp;
1992 	struct ucred *cred;
1993 	int igotlock;
1994 
1995 	/*
1996 	 * For the case that matters, this is the thread that set
1997 	 * MNTK_UNMOUNTF, so it will see it set. The code that follows is
1998 	 * done to ensure that any thread executing nfscl_getcl() after
1999 	 * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
2000 	 * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
2001 	 * explanation, courtesy of Alan Cox.
2002 	 * What follows is a snippet from Alan Cox's email at:
2003 	 * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
2004 	 *
2005 	 * 1. Set MNTK_UNMOUNTF
2006 	 * 2. Acquire a standard FreeBSD mutex "m".
2007 	 * 3. Update some data structures.
2008 	 * 4. Release mutex "m".
2009 	 *
2010 	 * Then, other threads that acquire "m" after step 4 has occurred will
2011 	 * see MNTK_UNMOUNTF as set.  But, other threads that beat thread X to
2012 	 * step 2 may or may not see MNTK_UNMOUNTF as set.
2013 	 */
2014 	NFSLOCKCLSTATE();
2015 	if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
2016 		fake_global++;
2017 		NFSUNLOCKCLSTATE();
2018 		NFSLOCKCLSTATE();
2019 	}
2020 
2021 	clp = nmp->nm_clp;
2022 	if (clp != NULL) {
2023 		if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
2024 			panic("nfscl umount");
2025 
2026 		/*
2027 		 * First, handshake with the nfscl renew thread, to terminate
2028 		 * it.
2029 		 */
2030 		clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
2031 		while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
2032 			(void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
2033 			    "nfsclumnt", hz);
2034 
2035 		/*
2036 		 * Now, get the exclusive lock on the client state, so
2037 		 * that no uses of the state are still in progress.
2038 		 */
2039 		do {
2040 			igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2041 			    NFSCLSTATEMUTEXPTR, NULL);
2042 		} while (!igotlock);
2043 		NFSUNLOCKCLSTATE();
2044 
2045 		/*
2046 		 * Free up all the state. It will expire on the server, but
2047 		 * maybe we should do a SetClientId/SetClientIdConfirm so
2048 		 * the server throws it away?
2049 		 */
2050 		LIST_REMOVE(clp, nfsc_list);
2051 		nfscl_delegreturnall(clp, p, dhp);
2052 		cred = newnfs_getcred();
2053 		if (NFSHASNFSV4N(nmp)) {
2054 			(void)nfsrpc_destroysession(nmp, clp, cred, p);
2055 			(void)nfsrpc_destroyclient(nmp, clp, cred, p);
2056 		} else
2057 			(void)nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2058 		nfscl_cleanclient(clp);
2059 		nmp->nm_clp = NULL;
2060 		NFSFREECRED(cred);
2061 		free(clp, M_NFSCLCLIENT);
2062 	} else
2063 		NFSUNLOCKCLSTATE();
2064 }
2065 
2066 /*
2067  * This function is called when a server replies with NFSERR_STALECLIENTID
2068  * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
2069  * doing Opens and Locks with reclaim. If these fail, it deletes the
2070  * corresponding state.
2071  */
2072 static void
2073 nfscl_recover(struct nfsclclient *clp, bool *retokp, struct ucred *cred,
2074     NFSPROC_T *p)
2075 {
2076 	struct nfsclowner *owp, *nowp;
2077 	struct nfsclopen *op, *nop;
2078 	struct nfscllockowner *lp, *nlp;
2079 	struct nfscllock *lop, *nlop;
2080 	struct nfscldeleg *dp, *ndp, *tdp;
2081 	struct nfsmount *nmp;
2082 	struct ucred *tcred;
2083 	struct nfsclopenhead extra_open;
2084 	struct nfscldeleghead extra_deleg;
2085 	struct nfsreq *rep;
2086 	u_int64_t len;
2087 	u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
2088 	int i, igotlock = 0, error, trycnt, firstlock;
2089 	struct nfscllayout *lyp, *nlyp;
2090 	bool recovered_one;
2091 
2092 	/*
2093 	 * First, lock the client structure, so everyone else will
2094 	 * block when trying to use state.
2095 	 */
2096 	NFSLOCKCLSTATE();
2097 	clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2098 	do {
2099 		igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2100 		    NFSCLSTATEMUTEXPTR, NULL);
2101 	} while (!igotlock);
2102 	NFSUNLOCKCLSTATE();
2103 
2104 	nmp = clp->nfsc_nmp;
2105 	if (nmp == NULL)
2106 		panic("nfscl recover");
2107 
2108 	/*
2109 	 * For now, just get rid of all layouts. There may be a need
2110 	 * to do LayoutCommit Ops with reclaim == true later.
2111 	 */
2112 	TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2113 		nfscl_freelayout(lyp);
2114 	TAILQ_INIT(&clp->nfsc_layout);
2115 	for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2116 		LIST_INIT(&clp->nfsc_layouthash[i]);
2117 
2118 	trycnt = 5;
2119 	tcred = NULL;
2120 	do {
2121 		error = nfsrpc_setclient(nmp, clp, 1, retokp, cred, p);
2122 	} while ((error == NFSERR_STALECLIENTID ||
2123 	     error == NFSERR_BADSESSION ||
2124 	     error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2125 	if (error) {
2126 		NFSLOCKCLSTATE();
2127 		clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2128 		    NFSCLFLAGS_RECVRINPROG);
2129 		wakeup(&clp->nfsc_flags);
2130 		nfsv4_unlock(&clp->nfsc_lock, 0);
2131 		NFSUNLOCKCLSTATE();
2132 		return;
2133 	}
2134 	clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2135 	clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2136 
2137 	/*
2138 	 * Mark requests already queued on the server, so that they don't
2139 	 * initiate another recovery cycle. Any requests already in the
2140 	 * queue that handle state information will have the old stale
2141 	 * clientid/stateid and will get a NFSERR_STALESTATEID,
2142 	 * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2143 	 * This will be translated to NFSERR_STALEDONTRECOVER when
2144 	 * R_DONTRECOVER is set.
2145 	 */
2146 	NFSLOCKREQ();
2147 	TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2148 		if (rep->r_nmp == nmp)
2149 			rep->r_flags |= R_DONTRECOVER;
2150 	}
2151 	NFSUNLOCKREQ();
2152 
2153 	/*
2154 	 * If nfsrpc_setclient() returns *retokp == true,
2155 	 * no more recovery is needed.
2156 	 */
2157 	if (*retokp)
2158 		goto out;
2159 
2160 	/*
2161 	 * Now, mark all delegations "need reclaim".
2162 	 */
2163 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2164 		dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2165 
2166 	TAILQ_INIT(&extra_deleg);
2167 	LIST_INIT(&extra_open);
2168 	/*
2169 	 * Now traverse the state lists, doing Open and Lock Reclaims.
2170 	 */
2171 	tcred = newnfs_getcred();
2172 	recovered_one = false;
2173 	owp = LIST_FIRST(&clp->nfsc_owner);
2174 	while (owp != NULL) {
2175 	    nowp = LIST_NEXT(owp, nfsow_list);
2176 	    owp->nfsow_seqid = 0;
2177 	    op = LIST_FIRST(&owp->nfsow_open);
2178 	    while (op != NULL) {
2179 		nop = LIST_NEXT(op, nfso_list);
2180 		if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2181 		    /* Search for a delegation to reclaim with the open */
2182 		    TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2183 			if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2184 			    continue;
2185 			if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2186 			    mode = NFSV4OPEN_ACCESSWRITE;
2187 			    delegtype = NFSV4OPEN_DELEGATEWRITE;
2188 			} else {
2189 			    mode = NFSV4OPEN_ACCESSREAD;
2190 			    delegtype = NFSV4OPEN_DELEGATEREAD;
2191 			}
2192 			if ((op->nfso_mode & mode) == mode &&
2193 			    op->nfso_fhlen == dp->nfsdl_fhlen &&
2194 			    !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2195 			    break;
2196 		    }
2197 		    ndp = dp;
2198 		    if (dp == NULL)
2199 			delegtype = NFSV4OPEN_DELEGATENONE;
2200 		    newnfs_copycred(&op->nfso_cred, tcred);
2201 		    error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2202 			op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2203 			op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2204 			tcred, p);
2205 		    if (!error) {
2206 			recovered_one = true;
2207 			/* Handle any replied delegation */
2208 			if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2209 			    || NFSMNT_RDONLY(nmp->nm_mountp))) {
2210 			    if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2211 				mode = NFSV4OPEN_ACCESSWRITE;
2212 			    else
2213 				mode = NFSV4OPEN_ACCESSREAD;
2214 			    TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2215 				if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2216 				    continue;
2217 				if ((op->nfso_mode & mode) == mode &&
2218 				    op->nfso_fhlen == dp->nfsdl_fhlen &&
2219 				    !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2220 				    op->nfso_fhlen)) {
2221 				    dp->nfsdl_stateid = ndp->nfsdl_stateid;
2222 				    dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2223 				    dp->nfsdl_ace = ndp->nfsdl_ace;
2224 				    dp->nfsdl_change = ndp->nfsdl_change;
2225 				    dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2226 				    if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2227 					dp->nfsdl_flags |= NFSCLDL_RECALL;
2228 				    free(ndp, M_NFSCLDELEG);
2229 				    ndp = NULL;
2230 				    break;
2231 				}
2232 			    }
2233 			}
2234 			if (ndp != NULL)
2235 			    TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2236 
2237 			/* and reclaim all byte range locks */
2238 			lp = LIST_FIRST(&op->nfso_lock);
2239 			while (lp != NULL) {
2240 			    nlp = LIST_NEXT(lp, nfsl_list);
2241 			    lp->nfsl_seqid = 0;
2242 			    firstlock = 1;
2243 			    lop = LIST_FIRST(&lp->nfsl_lock);
2244 			    while (lop != NULL) {
2245 				nlop = LIST_NEXT(lop, nfslo_list);
2246 				if (lop->nfslo_end == NFS64BITSSET)
2247 				    len = NFS64BITSSET;
2248 				else
2249 				    len = lop->nfslo_end - lop->nfslo_first;
2250 				error = nfscl_trylock(nmp, NULL,
2251 				    op->nfso_fh, op->nfso_fhlen, lp,
2252 				    firstlock, 1, lop->nfslo_first, len,
2253 				    lop->nfslo_type, tcred, p);
2254 				if (error != 0)
2255 				    nfscl_freelock(lop, 0);
2256 				else
2257 				    firstlock = 0;
2258 				lop = nlop;
2259 			    }
2260 			    /* If no locks, but a lockowner, just delete it. */
2261 			    if (LIST_EMPTY(&lp->nfsl_lock))
2262 				nfscl_freelockowner(lp, 0);
2263 			    lp = nlp;
2264 			}
2265 		    } else if (error == NFSERR_NOGRACE && !recovered_one &&
2266 			NFSHASNFSV4N(nmp)) {
2267 			/*
2268 			 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2269 			 * actually end up here, since the client will do
2270 			 * a recovery for NFSERR_BADSESSION, but will get
2271 			 * an NFSERR_NOGRACE reply for the first "reclaim"
2272 			 * attempt.
2273 			 * So, call nfscl_expireclient() to recover the
2274 			 * opens as best we can and then do a reclaim
2275 			 * complete and return.
2276 			 */
2277 			nfsrpc_reclaimcomplete(nmp, cred, p);
2278 			nfscl_expireclient(clp, nmp, tcred, p);
2279 			goto out;
2280 		    }
2281 		}
2282 		if (error != 0 && error != NFSERR_BADSESSION)
2283 		    nfscl_freeopen(op, 0, true);
2284 		op = nop;
2285 	    }
2286 	    owp = nowp;
2287 	}
2288 
2289 	/*
2290 	 * Now, try and get any delegations not yet reclaimed by cobbling
2291 	 * to-gether an appropriate open.
2292 	 */
2293 	nowp = NULL;
2294 	dp = TAILQ_FIRST(&clp->nfsc_deleg);
2295 	while (dp != NULL) {
2296 	    ndp = TAILQ_NEXT(dp, nfsdl_list);
2297 	    if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2298 		if (nowp == NULL) {
2299 		    nowp = malloc(
2300 			sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2301 		    /*
2302 		     * Name must be as long an largest possible
2303 		     * NFSV4CL_LOCKNAMELEN. 12 for now.
2304 		     */
2305 		    NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2306 			NFSV4CL_LOCKNAMELEN);
2307 		    LIST_INIT(&nowp->nfsow_open);
2308 		    nowp->nfsow_clp = clp;
2309 		    nowp->nfsow_seqid = 0;
2310 		    nowp->nfsow_defunct = 0;
2311 		    nfscl_lockinit(&nowp->nfsow_rwlock);
2312 		}
2313 		nop = NULL;
2314 		if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2315 		    nop = malloc(sizeof (struct nfsclopen) +
2316 			dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2317 		    nop->nfso_own = nowp;
2318 		    if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2319 			nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2320 			delegtype = NFSV4OPEN_DELEGATEWRITE;
2321 		    } else {
2322 			nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2323 			delegtype = NFSV4OPEN_DELEGATEREAD;
2324 		    }
2325 		    nop->nfso_opencnt = 0;
2326 		    nop->nfso_posixlock = 1;
2327 		    nop->nfso_fhlen = dp->nfsdl_fhlen;
2328 		    NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2329 		    LIST_INIT(&nop->nfso_lock);
2330 		    nop->nfso_stateid.seqid = 0;
2331 		    nop->nfso_stateid.other[0] = 0;
2332 		    nop->nfso_stateid.other[1] = 0;
2333 		    nop->nfso_stateid.other[2] = 0;
2334 		    newnfs_copycred(&dp->nfsdl_cred, tcred);
2335 		    newnfs_copyincred(tcred, &nop->nfso_cred);
2336 		    tdp = NULL;
2337 		    error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2338 			nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2339 			nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2340 			delegtype, tcred, p);
2341 		    if (tdp != NULL) {
2342 			if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2343 			    mode = NFSV4OPEN_ACCESSWRITE;
2344 			else
2345 			    mode = NFSV4OPEN_ACCESSREAD;
2346 			if ((nop->nfso_mode & mode) == mode &&
2347 			    nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2348 			    !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2349 			    nop->nfso_fhlen)) {
2350 			    dp->nfsdl_stateid = tdp->nfsdl_stateid;
2351 			    dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2352 			    dp->nfsdl_ace = tdp->nfsdl_ace;
2353 			    dp->nfsdl_change = tdp->nfsdl_change;
2354 			    dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2355 			    if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2356 				dp->nfsdl_flags |= NFSCLDL_RECALL;
2357 			    free(tdp, M_NFSCLDELEG);
2358 			} else {
2359 			    TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2360 			}
2361 		    }
2362 		}
2363 		if (error) {
2364 		    if (nop != NULL)
2365 			free(nop, M_NFSCLOPEN);
2366 		    if (error == NFSERR_NOGRACE && !recovered_one &&
2367 			NFSHASNFSV4N(nmp)) {
2368 			/*
2369 			 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2370 			 * actually end up here, since the client will do
2371 			 * a recovery for NFSERR_BADSESSION, but will get
2372 			 * an NFSERR_NOGRACE reply for the first "reclaim"
2373 			 * attempt.
2374 			 * So, call nfscl_expireclient() to recover the
2375 			 * opens as best we can and then do a reclaim
2376 			 * complete and return.
2377 			 */
2378 			nfsrpc_reclaimcomplete(nmp, cred, p);
2379 			nfscl_expireclient(clp, nmp, tcred, p);
2380 			free(nowp, M_NFSCLOWNER);
2381 			goto out;
2382 		    }
2383 		    /*
2384 		     * Couldn't reclaim it, so throw the state
2385 		     * away. Ouch!!
2386 		     */
2387 		    nfscl_cleandeleg(dp);
2388 		    nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
2389 		} else {
2390 		    recovered_one = true;
2391 		    LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2392 		}
2393 	    }
2394 	    dp = ndp;
2395 	}
2396 
2397 	/*
2398 	 * Now, get rid of extra Opens and Delegations.
2399 	 */
2400 	LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2401 		do {
2402 			newnfs_copycred(&op->nfso_cred, tcred);
2403 			error = nfscl_tryclose(op, tcred, nmp, p, true);
2404 			if (error == NFSERR_GRACE)
2405 				(void) nfs_catnap(PZERO, error, "nfsexcls");
2406 		} while (error == NFSERR_GRACE);
2407 		LIST_REMOVE(op, nfso_list);
2408 		free(op, M_NFSCLOPEN);
2409 	}
2410 	if (nowp != NULL)
2411 		free(nowp, M_NFSCLOWNER);
2412 
2413 	TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2414 		do {
2415 			newnfs_copycred(&dp->nfsdl_cred, tcred);
2416 			error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2417 			if (error == NFSERR_GRACE)
2418 				(void) nfs_catnap(PZERO, error, "nfsexdlg");
2419 		} while (error == NFSERR_GRACE);
2420 		TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2421 		free(dp, M_NFSCLDELEG);
2422 	}
2423 
2424 	/* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2425 	if (NFSHASNFSV4N(nmp))
2426 		(void)nfsrpc_reclaimcomplete(nmp, cred, p);
2427 
2428 out:
2429 	NFSLOCKCLSTATE();
2430 	clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2431 	wakeup(&clp->nfsc_flags);
2432 	nfsv4_unlock(&clp->nfsc_lock, 0);
2433 	NFSUNLOCKCLSTATE();
2434 	if (tcred != NULL)
2435 		NFSFREECRED(tcred);
2436 }
2437 
2438 /*
2439  * This function is called when a server replies with NFSERR_EXPIRED.
2440  * It deletes all state for the client and does a fresh SetClientId/confirm.
2441  * XXX Someday it should post a signal to the process(es) that hold the
2442  * state, so they know that lock state has been lost.
2443  */
2444 int
2445 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2446 {
2447 	struct nfsmount *nmp;
2448 	struct ucred *cred;
2449 	int igotlock = 0, error, trycnt;
2450 
2451 	/*
2452 	 * If the clientid has gone away or a new SetClientid has already
2453 	 * been done, just return ok.
2454 	 */
2455 	if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2456 		return (0);
2457 
2458 	/*
2459 	 * First, lock the client structure, so everyone else will
2460 	 * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2461 	 * that only one thread does the work.
2462 	 */
2463 	NFSLOCKCLSTATE();
2464 	clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2465 	do {
2466 		igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2467 		    NFSCLSTATEMUTEXPTR, NULL);
2468 	} while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2469 	if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2470 		if (igotlock)
2471 			nfsv4_unlock(&clp->nfsc_lock, 0);
2472 		NFSUNLOCKCLSTATE();
2473 		return (0);
2474 	}
2475 	clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2476 	NFSUNLOCKCLSTATE();
2477 
2478 	nmp = clp->nfsc_nmp;
2479 	if (nmp == NULL)
2480 		panic("nfscl expired");
2481 	cred = newnfs_getcred();
2482 	trycnt = 5;
2483 	do {
2484 		error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2485 	} while ((error == NFSERR_STALECLIENTID ||
2486 	     error == NFSERR_BADSESSION ||
2487 	     error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2488 	if (error) {
2489 		NFSLOCKCLSTATE();
2490 		clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2491 	} else {
2492 		/*
2493 		 * Expire the state for the client.
2494 		 */
2495 		nfscl_expireclient(clp, nmp, cred, p);
2496 		NFSLOCKCLSTATE();
2497 		clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2498 		clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2499 	}
2500 	clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2501 	wakeup(&clp->nfsc_flags);
2502 	nfsv4_unlock(&clp->nfsc_lock, 0);
2503 	NFSUNLOCKCLSTATE();
2504 	NFSFREECRED(cred);
2505 	return (error);
2506 }
2507 
2508 /*
2509  * This function inserts a lock in the list after insert_lop.
2510  */
2511 static void
2512 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2513     struct nfscllock *insert_lop, int local)
2514 {
2515 
2516 	if ((struct nfscllockowner *)insert_lop == lp)
2517 		LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2518 	else
2519 		LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2520 	if (local)
2521 		nfsstatsv1.cllocallocks++;
2522 	else
2523 		nfsstatsv1.cllocks++;
2524 }
2525 
2526 /*
2527  * This function updates the locking for a lock owner and given file. It
2528  * maintains a list of lock ranges ordered on increasing file offset that
2529  * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2530  * It always adds new_lop to the list and sometimes uses the one pointed
2531  * at by other_lopp.
2532  * Returns 1 if the locks were modified, 0 otherwise.
2533  */
2534 static int
2535 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2536     struct nfscllock **other_lopp, int local)
2537 {
2538 	struct nfscllock *new_lop = *new_lopp;
2539 	struct nfscllock *lop, *tlop, *ilop;
2540 	struct nfscllock *other_lop;
2541 	int unlock = 0, modified = 0;
2542 	u_int64_t tmp;
2543 
2544 	/*
2545 	 * Work down the list until the lock is merged.
2546 	 */
2547 	if (new_lop->nfslo_type == F_UNLCK)
2548 		unlock = 1;
2549 	ilop = (struct nfscllock *)lp;
2550 	lop = LIST_FIRST(&lp->nfsl_lock);
2551 	while (lop != NULL) {
2552 	    /*
2553 	     * Only check locks for this file that aren't before the start of
2554 	     * new lock's range.
2555 	     */
2556 	    if (lop->nfslo_end >= new_lop->nfslo_first) {
2557 		if (new_lop->nfslo_end < lop->nfslo_first) {
2558 		    /*
2559 		     * If the new lock ends before the start of the
2560 		     * current lock's range, no merge, just insert
2561 		     * the new lock.
2562 		     */
2563 		    break;
2564 		}
2565 		if (new_lop->nfslo_type == lop->nfslo_type ||
2566 		    (new_lop->nfslo_first <= lop->nfslo_first &&
2567 		     new_lop->nfslo_end >= lop->nfslo_end)) {
2568 		    /*
2569 		     * This lock can be absorbed by the new lock/unlock.
2570 		     * This happens when it covers the entire range
2571 		     * of the old lock or is contiguous
2572 		     * with the old lock and is of the same type or an
2573 		     * unlock.
2574 		     */
2575 		    if (new_lop->nfslo_type != lop->nfslo_type ||
2576 			new_lop->nfslo_first != lop->nfslo_first ||
2577 			new_lop->nfslo_end != lop->nfslo_end)
2578 			modified = 1;
2579 		    if (lop->nfslo_first < new_lop->nfslo_first)
2580 			new_lop->nfslo_first = lop->nfslo_first;
2581 		    if (lop->nfslo_end > new_lop->nfslo_end)
2582 			new_lop->nfslo_end = lop->nfslo_end;
2583 		    tlop = lop;
2584 		    lop = LIST_NEXT(lop, nfslo_list);
2585 		    nfscl_freelock(tlop, local);
2586 		    continue;
2587 		}
2588 
2589 		/*
2590 		 * All these cases are for contiguous locks that are not the
2591 		 * same type, so they can't be merged.
2592 		 */
2593 		if (new_lop->nfslo_first <= lop->nfslo_first) {
2594 		    /*
2595 		     * This case is where the new lock overlaps with the
2596 		     * first part of the old lock. Move the start of the
2597 		     * old lock to just past the end of the new lock. The
2598 		     * new lock will be inserted in front of the old, since
2599 		     * ilop hasn't been updated. (We are done now.)
2600 		     */
2601 		    if (lop->nfslo_first != new_lop->nfslo_end) {
2602 			lop->nfslo_first = new_lop->nfslo_end;
2603 			modified = 1;
2604 		    }
2605 		    break;
2606 		}
2607 		if (new_lop->nfslo_end >= lop->nfslo_end) {
2608 		    /*
2609 		     * This case is where the new lock overlaps with the
2610 		     * end of the old lock's range. Move the old lock's
2611 		     * end to just before the new lock's first and insert
2612 		     * the new lock after the old lock.
2613 		     * Might not be done yet, since the new lock could
2614 		     * overlap further locks with higher ranges.
2615 		     */
2616 		    if (lop->nfslo_end != new_lop->nfslo_first) {
2617 			lop->nfslo_end = new_lop->nfslo_first;
2618 			modified = 1;
2619 		    }
2620 		    ilop = lop;
2621 		    lop = LIST_NEXT(lop, nfslo_list);
2622 		    continue;
2623 		}
2624 		/*
2625 		 * The final case is where the new lock's range is in the
2626 		 * middle of the current lock's and splits the current lock
2627 		 * up. Use *other_lopp to handle the second part of the
2628 		 * split old lock range. (We are done now.)
2629 		 * For unlock, we use new_lop as other_lop and tmp, since
2630 		 * other_lop and new_lop are the same for this case.
2631 		 * We noted the unlock case above, so we don't need
2632 		 * new_lop->nfslo_type any longer.
2633 		 */
2634 		tmp = new_lop->nfslo_first;
2635 		if (unlock) {
2636 		    other_lop = new_lop;
2637 		    *new_lopp = NULL;
2638 		} else {
2639 		    other_lop = *other_lopp;
2640 		    *other_lopp = NULL;
2641 		}
2642 		other_lop->nfslo_first = new_lop->nfslo_end;
2643 		other_lop->nfslo_end = lop->nfslo_end;
2644 		other_lop->nfslo_type = lop->nfslo_type;
2645 		lop->nfslo_end = tmp;
2646 		nfscl_insertlock(lp, other_lop, lop, local);
2647 		ilop = lop;
2648 		modified = 1;
2649 		break;
2650 	    }
2651 	    ilop = lop;
2652 	    lop = LIST_NEXT(lop, nfslo_list);
2653 	    if (lop == NULL)
2654 		break;
2655 	}
2656 
2657 	/*
2658 	 * Insert the new lock in the list at the appropriate place.
2659 	 */
2660 	if (!unlock) {
2661 		nfscl_insertlock(lp, new_lop, ilop, local);
2662 		*new_lopp = NULL;
2663 		modified = 1;
2664 	}
2665 	return (modified);
2666 }
2667 
2668 /*
2669  * This function must be run as a kernel thread.
2670  * It does Renew Ops and recovery, when required.
2671  */
2672 void
2673 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2674 {
2675 	struct nfsclowner *owp, *nowp;
2676 	struct nfsclopen *op;
2677 	struct nfscllockowner *lp, *nlp;
2678 	struct nfscldeleghead dh;
2679 	struct nfscldeleg *dp, *ndp;
2680 	struct ucred *cred;
2681 	u_int32_t clidrev;
2682 	int error, cbpathdown, islept, igotlock, ret, clearok;
2683 	uint32_t recover_done_time = 0;
2684 	time_t mytime;
2685 	static time_t prevsec = 0;
2686 	struct nfscllockownerfh *lfhp, *nlfhp;
2687 	struct nfscllockownerfhhead lfh;
2688 	struct nfscllayout *lyp, *nlyp;
2689 	struct nfscldevinfo *dip, *ndip;
2690 	struct nfscllayouthead rlh;
2691 	struct nfsclrecalllayout *recallp;
2692 	struct nfsclds *dsp;
2693 	bool retok;
2694 	struct mount *mp;
2695 	vnode_t vp;
2696 
2697 	cred = newnfs_getcred();
2698 	NFSLOCKCLSTATE();
2699 	clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2700 	mp = clp->nfsc_nmp->nm_mountp;
2701 	NFSUNLOCKCLSTATE();
2702 	for(;;) {
2703 		newnfs_setroot(cred);
2704 		cbpathdown = 0;
2705 		if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2706 			/*
2707 			 * Only allow one full recover within 1/2 of the lease
2708 			 * duration (nfsc_renew).
2709 			 * retok is value/result.  If passed in set to true,
2710 			 * it indicates only a CreateSession operation should
2711 			 * be attempted.
2712 			 * If it is returned true, it indicates that the
2713 			 * recovery only required a CreateSession.
2714 			 */
2715 			retok = true;
2716 			if (recover_done_time < NFSD_MONOSEC) {
2717 				recover_done_time = NFSD_MONOSEC +
2718 				    clp->nfsc_renew;
2719 				retok = false;
2720 			}
2721 			NFSCL_DEBUG(1, "Doing recovery, only "
2722 			    "createsession=%d\n", retok);
2723 			nfscl_recover(clp, &retok, cred, p);
2724 		}
2725 		if (clp->nfsc_expire <= NFSD_MONOSEC &&
2726 		    (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2727 			clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2728 			clidrev = clp->nfsc_clientidrev;
2729 			error = nfsrpc_renew(clp, NULL, cred, p);
2730 			if (error == NFSERR_CBPATHDOWN)
2731 			    cbpathdown = 1;
2732 			else if (error == NFSERR_STALECLIENTID ||
2733 			    error == NFSERR_BADSESSION) {
2734 			    NFSLOCKCLSTATE();
2735 			    clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2736 			    NFSUNLOCKCLSTATE();
2737 			} else if (error == NFSERR_EXPIRED)
2738 			    (void) nfscl_hasexpired(clp, clidrev, p);
2739 		}
2740 
2741 checkdsrenew:
2742 		if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2743 			/* Do renews for any DS sessions. */
2744 			NFSLOCKMNT(clp->nfsc_nmp);
2745 			/* Skip first entry, since the MDS is handled above. */
2746 			dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2747 			if (dsp != NULL)
2748 				dsp = TAILQ_NEXT(dsp, nfsclds_list);
2749 			while (dsp != NULL) {
2750 				if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2751 				    dsp->nfsclds_sess.nfsess_defunct == 0) {
2752 					dsp->nfsclds_expire = NFSD_MONOSEC +
2753 					    clp->nfsc_renew;
2754 					NFSUNLOCKMNT(clp->nfsc_nmp);
2755 					(void)nfsrpc_renew(clp, dsp, cred, p);
2756 					goto checkdsrenew;
2757 				}
2758 				dsp = TAILQ_NEXT(dsp, nfsclds_list);
2759 			}
2760 			NFSUNLOCKMNT(clp->nfsc_nmp);
2761 		}
2762 
2763 		TAILQ_INIT(&dh);
2764 		NFSLOCKCLSTATE();
2765 		if (cbpathdown)
2766 			/* It's a Total Recall! */
2767 			nfscl_totalrecall(clp);
2768 
2769 		/*
2770 		 * Now, handle defunct owners.
2771 		 */
2772 		LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2773 			if (LIST_EMPTY(&owp->nfsow_open)) {
2774 				if (owp->nfsow_defunct != 0)
2775 					nfscl_freeopenowner(owp, 0);
2776 			}
2777 		}
2778 
2779 		/*
2780 		 * Do the recall on any delegations. To avoid trouble, always
2781 		 * come back up here after having slept.
2782 		 */
2783 		igotlock = 0;
2784 tryagain:
2785 		dp = TAILQ_FIRST(&clp->nfsc_deleg);
2786 		while (dp != NULL) {
2787 			ndp = TAILQ_NEXT(dp, nfsdl_list);
2788 			if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2789 				/*
2790 				 * Wait for outstanding I/O ops to be done.
2791 				 */
2792 				if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2793 				    if (igotlock) {
2794 					nfsv4_unlock(&clp->nfsc_lock, 0);
2795 					igotlock = 0;
2796 				    }
2797 				    dp->nfsdl_rwlock.nfslock_lock |=
2798 					NFSV4LOCK_WANTED;
2799 				    msleep(&dp->nfsdl_rwlock,
2800 					NFSCLSTATEMUTEXPTR, PVFS, "nfscld",
2801 					5 * hz);
2802 				    if (NFSCL_FORCEDISM(mp))
2803 					goto terminate;
2804 				    goto tryagain;
2805 				}
2806 				while (!igotlock) {
2807 				    igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2808 					&islept, NFSCLSTATEMUTEXPTR, mp);
2809 				    if (igotlock == 0 && NFSCL_FORCEDISM(mp))
2810 					goto terminate;
2811 				    if (islept)
2812 					goto tryagain;
2813 				}
2814 				NFSUNLOCKCLSTATE();
2815 				newnfs_copycred(&dp->nfsdl_cred, cred);
2816 				ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2817 				    NULL, cred, p, 1, &vp);
2818 				if (!ret) {
2819 				    nfscl_cleandeleg(dp);
2820 				    TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2821 					nfsdl_list);
2822 				    LIST_REMOVE(dp, nfsdl_hash);
2823 				    TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2824 				    nfscl_delegcnt--;
2825 				    nfsstatsv1.cldelegates--;
2826 				}
2827 				NFSLOCKCLSTATE();
2828 				/*
2829 				 * The nfsc_lock must be released before doing
2830 				 * vrele(), since it might call nfs_inactive().
2831 				 * For the unlikely case where the vnode failed
2832 				 * to be acquired by nfscl_recalldeleg(), a
2833 				 * VOP_RECLAIM() should be in progress and it
2834 				 * will return the delegation.
2835 				 */
2836 				nfsv4_unlock(&clp->nfsc_lock, 0);
2837 				igotlock = 0;
2838 				if (vp != NULL) {
2839 					NFSUNLOCKCLSTATE();
2840 					vrele(vp);
2841 					NFSLOCKCLSTATE();
2842 				}
2843 				goto tryagain;
2844 			}
2845 			dp = ndp;
2846 		}
2847 
2848 		/*
2849 		 * Clear out old delegations, if we are above the high water
2850 		 * mark. Only clear out ones with no state related to them.
2851 		 * The tailq list is in LRU order.
2852 		 */
2853 		dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2854 		while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2855 		    ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2856 		    if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2857 			dp->nfsdl_rwlock.nfslock_lock == 0 &&
2858 			dp->nfsdl_timestamp < NFSD_MONOSEC &&
2859 			(dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2860 			  NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2861 			clearok = 1;
2862 			LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2863 			    op = LIST_FIRST(&owp->nfsow_open);
2864 			    if (op != NULL) {
2865 				clearok = 0;
2866 				break;
2867 			    }
2868 			}
2869 			if (clearok) {
2870 			    LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2871 				if (!LIST_EMPTY(&lp->nfsl_lock)) {
2872 				    clearok = 0;
2873 				    break;
2874 				}
2875 			    }
2876 			}
2877 			if (clearok) {
2878 			    TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2879 			    LIST_REMOVE(dp, nfsdl_hash);
2880 			    TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2881 			    nfscl_delegcnt--;
2882 			    nfsstatsv1.cldelegates--;
2883 			}
2884 		    }
2885 		    dp = ndp;
2886 		}
2887 		if (igotlock)
2888 			nfsv4_unlock(&clp->nfsc_lock, 0);
2889 
2890 		/*
2891 		 * Do the recall on any layouts. To avoid trouble, always
2892 		 * come back up here after having slept.
2893 		 */
2894 		TAILQ_INIT(&rlh);
2895 tryagain2:
2896 		TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2897 			if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2898 				/*
2899 				 * Wait for outstanding I/O ops to be done.
2900 				 */
2901 				if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2902 				    (lyp->nfsly_lock.nfslock_lock &
2903 				     NFSV4LOCK_LOCK) != 0) {
2904 					lyp->nfsly_lock.nfslock_lock |=
2905 					    NFSV4LOCK_WANTED;
2906 					msleep(&lyp->nfsly_lock.nfslock_lock,
2907 					    NFSCLSTATEMUTEXPTR, PVFS, "nfslyp",
2908 					    5 * hz);
2909 					if (NFSCL_FORCEDISM(mp))
2910 					    goto terminate;
2911 					goto tryagain2;
2912 				}
2913 				/* Move the layout to the recall list. */
2914 				TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2915 				    nfsly_list);
2916 				LIST_REMOVE(lyp, nfsly_hash);
2917 				TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2918 
2919 				/* Handle any layout commits. */
2920 				if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2921 				    (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2922 					lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2923 					NFSUNLOCKCLSTATE();
2924 					NFSCL_DEBUG(3, "do layoutcommit\n");
2925 					nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2926 					    cred, p);
2927 					NFSLOCKCLSTATE();
2928 					goto tryagain2;
2929 				}
2930 			}
2931 		}
2932 
2933 		/* Now, look for stale layouts. */
2934 		lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2935 		while (lyp != NULL) {
2936 			nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2937 			if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2938 			    (lyp->nfsly_flags & (NFSLY_RECALL |
2939 			     NFSLY_RETONCLOSE)) == 0 &&
2940 			    lyp->nfsly_lock.nfslock_usecnt == 0 &&
2941 			    lyp->nfsly_lock.nfslock_lock == 0) {
2942 				NFSCL_DEBUG(4, "ret stale lay=%d\n",
2943 				    nfscl_layoutcnt);
2944 				recallp = malloc(sizeof(*recallp),
2945 				    M_NFSLAYRECALL, M_NOWAIT);
2946 				if (recallp == NULL)
2947 					break;
2948 				(void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2949 				    lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2950 				    lyp->nfsly_stateid.seqid, 0, 0, NULL,
2951 				    recallp);
2952 			}
2953 			lyp = nlyp;
2954 		}
2955 
2956 		/*
2957 		 * Free up any unreferenced device info structures.
2958 		 */
2959 		LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2960 			if (dip->nfsdi_layoutrefs == 0 &&
2961 			    dip->nfsdi_refcnt == 0) {
2962 				NFSCL_DEBUG(4, "freeing devinfo\n");
2963 				LIST_REMOVE(dip, nfsdi_list);
2964 				nfscl_freedevinfo(dip);
2965 			}
2966 		}
2967 		NFSUNLOCKCLSTATE();
2968 
2969 		/* Do layout return(s), as required. */
2970 		TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2971 			TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2972 			NFSCL_DEBUG(4, "ret layout\n");
2973 			nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2974 			if ((lyp->nfsly_flags & NFSLY_RETONCLOSE) != 0) {
2975 				NFSLOCKCLSTATE();
2976 				lyp->nfsly_flags |= NFSLY_RETURNED;
2977 				wakeup(lyp);
2978 				NFSUNLOCKCLSTATE();
2979 			} else
2980 				nfscl_freelayout(lyp);
2981 		}
2982 
2983 		/*
2984 		 * Delegreturn any delegations cleaned out or recalled.
2985 		 */
2986 		TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2987 			newnfs_copycred(&dp->nfsdl_cred, cred);
2988 			(void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2989 			TAILQ_REMOVE(&dh, dp, nfsdl_list);
2990 			free(dp, M_NFSCLDELEG);
2991 		}
2992 
2993 		SLIST_INIT(&lfh);
2994 		/*
2995 		 * Call nfscl_cleanupkext() once per second to check for
2996 		 * open/lock owners where the process has exited.
2997 		 */
2998 		mytime = NFSD_MONOSEC;
2999 		if (prevsec != mytime) {
3000 			prevsec = mytime;
3001 			nfscl_cleanupkext(clp, &lfh);
3002 		}
3003 
3004 		/*
3005 		 * Do a ReleaseLockOwner for all lock owners where the
3006 		 * associated process no longer exists, as found by
3007 		 * nfscl_cleanupkext().
3008 		 */
3009 		newnfs_setroot(cred);
3010 		SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
3011 			LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
3012 			    nlp) {
3013 				(void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
3014 				    lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
3015 				    p);
3016 				nfscl_freelockowner(lp, 0);
3017 			}
3018 			free(lfhp, M_TEMP);
3019 		}
3020 		SLIST_INIT(&lfh);
3021 
3022 		NFSLOCKCLSTATE();
3023 		if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
3024 			(void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
3025 			    hz);
3026 terminate:
3027 		if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
3028 			clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
3029 			NFSUNLOCKCLSTATE();
3030 			NFSFREECRED(cred);
3031 			wakeup((caddr_t)clp);
3032 			return;
3033 		}
3034 		NFSUNLOCKCLSTATE();
3035 	}
3036 }
3037 
3038 /*
3039  * Initiate state recovery. Called when NFSERR_STALECLIENTID,
3040  * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
3041  */
3042 void
3043 nfscl_initiate_recovery(struct nfsclclient *clp)
3044 {
3045 
3046 	if (clp == NULL)
3047 		return;
3048 	NFSLOCKCLSTATE();
3049 	clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
3050 	NFSUNLOCKCLSTATE();
3051 	wakeup((caddr_t)clp);
3052 }
3053 
3054 /*
3055  * Dump out the state stuff for debugging.
3056  */
3057 void
3058 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
3059     int lockowner, int locks)
3060 {
3061 	struct nfsclclient *clp;
3062 	struct nfsclowner *owp;
3063 	struct nfsclopen *op;
3064 	struct nfscllockowner *lp;
3065 	struct nfscllock *lop;
3066 	struct nfscldeleg *dp;
3067 
3068 	clp = nmp->nm_clp;
3069 	if (clp == NULL) {
3070 		printf("nfscl dumpstate NULL clp\n");
3071 		return;
3072 	}
3073 	NFSLOCKCLSTATE();
3074 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
3075 	  LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3076 	    if (openowner && !LIST_EMPTY(&owp->nfsow_open))
3077 		printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
3078 		    owp->nfsow_owner[0], owp->nfsow_owner[1],
3079 		    owp->nfsow_owner[2], owp->nfsow_owner[3],
3080 		    owp->nfsow_seqid);
3081 	    LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3082 		if (opens)
3083 		    printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
3084 			op->nfso_stateid.other[0], op->nfso_stateid.other[1],
3085 			op->nfso_stateid.other[2], op->nfso_opencnt,
3086 			op->nfso_fh[12]);
3087 		LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
3088 		    if (lockowner)
3089 			printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3090 			    lp->nfsl_owner[0], lp->nfsl_owner[1],
3091 			    lp->nfsl_owner[2], lp->nfsl_owner[3],
3092 			    lp->nfsl_seqid,
3093 			    lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3094 			    lp->nfsl_stateid.other[2]);
3095 		    LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3096 			if (locks)
3097 #ifdef __FreeBSD__
3098 			    printf("lck typ=%d fst=%ju end=%ju\n",
3099 				lop->nfslo_type, (intmax_t)lop->nfslo_first,
3100 				(intmax_t)lop->nfslo_end);
3101 #else
3102 			    printf("lck typ=%d fst=%qd end=%qd\n",
3103 				lop->nfslo_type, lop->nfslo_first,
3104 				lop->nfslo_end);
3105 #endif
3106 		    }
3107 		}
3108 	    }
3109 	  }
3110 	}
3111 	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3112 	    if (openowner && !LIST_EMPTY(&owp->nfsow_open))
3113 		printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
3114 		    owp->nfsow_owner[0], owp->nfsow_owner[1],
3115 		    owp->nfsow_owner[2], owp->nfsow_owner[3],
3116 		    owp->nfsow_seqid);
3117 	    LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3118 		if (opens)
3119 		    printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
3120 			op->nfso_stateid.other[0], op->nfso_stateid.other[1],
3121 			op->nfso_stateid.other[2], op->nfso_opencnt,
3122 			op->nfso_fh[12]);
3123 		LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
3124 		    if (lockowner)
3125 			printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3126 			    lp->nfsl_owner[0], lp->nfsl_owner[1],
3127 			    lp->nfsl_owner[2], lp->nfsl_owner[3],
3128 			    lp->nfsl_seqid,
3129 			    lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3130 			    lp->nfsl_stateid.other[2]);
3131 		    LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3132 			if (locks)
3133 #ifdef __FreeBSD__
3134 			    printf("lck typ=%d fst=%ju end=%ju\n",
3135 				lop->nfslo_type, (intmax_t)lop->nfslo_first,
3136 				(intmax_t)lop->nfslo_end);
3137 #else
3138 			    printf("lck typ=%d fst=%qd end=%qd\n",
3139 				lop->nfslo_type, lop->nfslo_first,
3140 				lop->nfslo_end);
3141 #endif
3142 		    }
3143 		}
3144 	    }
3145 	}
3146 	NFSUNLOCKCLSTATE();
3147 }
3148 
3149 /*
3150  * Check for duplicate open owners and opens.
3151  * (Only used as a diagnostic aid.)
3152  */
3153 void
3154 nfscl_dupopen(vnode_t vp, int dupopens)
3155 {
3156 	struct nfsclclient *clp;
3157 	struct nfsclowner *owp, *owp2;
3158 	struct nfsclopen *op, *op2;
3159 	struct nfsfh *nfhp;
3160 
3161 	clp = VFSTONFS(vp->v_mount)->nm_clp;
3162 	if (clp == NULL) {
3163 		printf("nfscl dupopen NULL clp\n");
3164 		return;
3165 	}
3166 	nfhp = VTONFS(vp)->n_fhp;
3167 	NFSLOCKCLSTATE();
3168 
3169 	/*
3170 	 * First, search for duplicate owners.
3171 	 * These should never happen!
3172 	 */
3173 	LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3174 	    LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3175 		if (owp != owp2 &&
3176 		    !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
3177 		    NFSV4CL_LOCKNAMELEN)) {
3178 			NFSUNLOCKCLSTATE();
3179 			printf("DUP OWNER\n");
3180 			nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3181 			return;
3182 		}
3183 	    }
3184 	}
3185 
3186 	/*
3187 	 * Now, search for duplicate stateids.
3188 	 * These shouldn't happen, either.
3189 	 */
3190 	LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3191 	    LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3192 		LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3193 		    LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3194 			if (op != op2 &&
3195 			    (op->nfso_stateid.other[0] != 0 ||
3196 			     op->nfso_stateid.other[1] != 0 ||
3197 			     op->nfso_stateid.other[2] != 0) &&
3198 			    op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3199 			    op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3200 			    op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3201 			    NFSUNLOCKCLSTATE();
3202 			    printf("DUP STATEID\n");
3203 			    nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3204 			    return;
3205 			}
3206 		    }
3207 		}
3208 	    }
3209 	}
3210 
3211 	/*
3212 	 * Now search for duplicate opens.
3213 	 * Duplicate opens for the same owner
3214 	 * should never occur. Other duplicates are
3215 	 * possible and are checked for if "dupopens"
3216 	 * is true.
3217 	 */
3218 	LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3219 	    LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3220 		if (nfhp->nfh_len == op2->nfso_fhlen &&
3221 		    !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3222 		    LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3223 			LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3224 			    if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3225 				!NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3226 				(!NFSBCMP(op->nfso_own->nfsow_owner,
3227 				 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3228 				 dupopens)) {
3229 				if (!NFSBCMP(op->nfso_own->nfsow_owner,
3230 				    op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3231 				    NFSUNLOCKCLSTATE();
3232 				    printf("BADDUP OPEN\n");
3233 				} else {
3234 				    NFSUNLOCKCLSTATE();
3235 				    printf("DUP OPEN\n");
3236 				}
3237 				nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0,
3238 				    0);
3239 				return;
3240 			    }
3241 			}
3242 		    }
3243 		}
3244 	    }
3245 	}
3246 	NFSUNLOCKCLSTATE();
3247 }
3248 
3249 /*
3250  * During close, find an open that needs to be dereferenced and
3251  * dereference it. If there are no more opens for this file,
3252  * log a message to that effect.
3253  * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3254  * on the file's vnode.
3255  * This is the safe way, since it is difficult to identify
3256  * which open the close is for and I/O can be performed after the
3257  * close(2) system call when a file is mmap'd.
3258  * If it returns 0 for success, there will be a referenced
3259  * clp returned via clpp.
3260  */
3261 int
3262 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3263 {
3264 	struct nfsclclient *clp;
3265 	struct nfsclowner *owp;
3266 	struct nfsclopen *op;
3267 	struct nfscldeleg *dp;
3268 	struct nfsfh *nfhp;
3269 	int error, notdecr;
3270 
3271 	error = nfscl_getcl(vp->v_mount, NULL, NULL, false, true, &clp);
3272 	if (error)
3273 		return (error);
3274 	*clpp = clp;
3275 
3276 	nfhp = VTONFS(vp)->n_fhp;
3277 	notdecr = 1;
3278 	NFSLOCKCLSTATE();
3279 	/*
3280 	 * First, look for one under a delegation that was locally issued
3281 	 * and just decrement the opencnt for it. Since all my Opens against
3282 	 * the server are DENY_NONE, I don't see a problem with hanging
3283 	 * onto them. (It is much easier to use one of the extant Opens
3284 	 * that I already have on the server when a Delegation is recalled
3285 	 * than to do fresh Opens.) Someday, I might need to rethink this, but.
3286 	 */
3287 	dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3288 	if (dp != NULL) {
3289 		LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3290 			op = LIST_FIRST(&owp->nfsow_open);
3291 			if (op != NULL) {
3292 				/*
3293 				 * Since a delegation is for a file, there
3294 				 * should never be more than one open for
3295 				 * each openowner.
3296 				 */
3297 				if (LIST_NEXT(op, nfso_list) != NULL)
3298 					panic("nfscdeleg opens");
3299 				if (notdecr && op->nfso_opencnt > 0) {
3300 					notdecr = 0;
3301 					op->nfso_opencnt--;
3302 					break;
3303 				}
3304 			}
3305 		}
3306 	}
3307 
3308 	/* Now process the opens against the server. */
3309 	LIST_FOREACH(op, NFSCLOPENHASH(clp, nfhp->nfh_fh, nfhp->nfh_len),
3310 	    nfso_hash) {
3311 		if (op->nfso_fhlen == nfhp->nfh_len &&
3312 		    !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3313 		    nfhp->nfh_len)) {
3314 			/* Found an open, decrement cnt if possible */
3315 			if (notdecr && op->nfso_opencnt > 0) {
3316 				notdecr = 0;
3317 				op->nfso_opencnt--;
3318 			}
3319 			/*
3320 			 * There are more opens, so just return.
3321 			 */
3322 			if (op->nfso_opencnt > 0) {
3323 				NFSUNLOCKCLSTATE();
3324 				return (0);
3325 			}
3326 		}
3327 	}
3328 	NFSUNLOCKCLSTATE();
3329 	if (notdecr)
3330 		printf("nfscl: never fnd open\n");
3331 	return (0);
3332 }
3333 
3334 int
3335 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3336 {
3337 	struct nfsclclient *clp;
3338 	struct nfsmount *nmp;
3339 	struct nfsclowner *owp, *nowp;
3340 	struct nfsclopen *op, *nop;
3341 	struct nfsclopenhead delayed;
3342 	struct nfscldeleg *dp;
3343 	struct nfsfh *nfhp;
3344 	struct nfsclrecalllayout *recallp;
3345 	struct nfscllayout *lyp;
3346 	int error;
3347 
3348 	error = nfscl_getcl(vp->v_mount, NULL, NULL, false, true, &clp);
3349 	if (error)
3350 		return (error);
3351 	*clpp = clp;
3352 
3353 	nmp = VFSTONFS(vp->v_mount);
3354 	nfhp = VTONFS(vp)->n_fhp;
3355 	recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3356 	NFSLOCKCLSTATE();
3357 	/*
3358 	 * First get rid of the local Open structures, which should be no
3359 	 * longer in use.
3360 	 */
3361 	dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3362 	if (dp != NULL) {
3363 		LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3364 			op = LIST_FIRST(&owp->nfsow_open);
3365 			if (op != NULL) {
3366 				KASSERT((op->nfso_opencnt == 0),
3367 				    ("nfscl: bad open cnt on deleg"));
3368 				nfscl_freeopen(op, 1, true);
3369 			}
3370 			nfscl_freeopenowner(owp, 1);
3371 		}
3372 	}
3373 
3374 	/* Return any layouts marked return on close. */
3375 	nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp,
3376 	    &lyp);
3377 
3378 	/* Now process the opens against the server. */
3379 	LIST_INIT(&delayed);
3380 lookformore:
3381 	LIST_FOREACH(op, NFSCLOPENHASH(clp, nfhp->nfh_fh, nfhp->nfh_len),
3382 	    nfso_hash) {
3383 		if (op->nfso_fhlen == nfhp->nfh_len &&
3384 		    !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3385 		    nfhp->nfh_len)) {
3386 			/* Found an open, close it. */
3387 #ifdef DIAGNOSTIC
3388 			KASSERT((op->nfso_opencnt == 0),
3389 			    ("nfscl: bad open cnt on server (%d)",
3390 			     op->nfso_opencnt));
3391 #endif
3392 			NFSUNLOCKCLSTATE();
3393 			if (NFSHASNFSV4N(nmp))
3394 				error = nfsrpc_doclose(nmp, op, p, false, true);
3395 			else
3396 				error = nfsrpc_doclose(nmp, op, p, true, true);
3397 			NFSLOCKCLSTATE();
3398 			if (error == NFSERR_DELAY) {
3399 				nfscl_unlinkopen(op);
3400 				op->nfso_own = NULL;
3401 				LIST_INSERT_HEAD(&delayed, op, nfso_list);
3402 			}
3403 			goto lookformore;
3404 		}
3405 	}
3406 	nfscl_clrelease(clp);
3407 
3408 	/* Now, wait for any layout that is returned upon close. */
3409 	if (lyp != NULL) {
3410 		while ((lyp->nfsly_flags & NFSLY_RETURNED) == 0) {
3411 			if (NFSCL_FORCEDISM(nmp->nm_mountp)) {
3412 				lyp = NULL;
3413 				break;
3414 			}
3415 			msleep(lyp, NFSCLSTATEMUTEXPTR, PZERO, "nfslroc", hz);
3416 		}
3417 		if (lyp != NULL)
3418 			nfscl_freelayout(lyp);
3419 	}
3420 
3421 	NFSUNLOCKCLSTATE();
3422 	/*
3423 	 * recallp has been set NULL by nfscl_retoncloselayout() if it was
3424 	 * used by the function, but calling free() with a NULL pointer is ok.
3425 	 */
3426 	free(recallp, M_NFSLAYRECALL);
3427 
3428 	/* Now, loop retrying the delayed closes. */
3429 	LIST_FOREACH_SAFE(op, &delayed, nfso_list, nop) {
3430 		nfsrpc_doclose(nmp, op, p, true, false);
3431 		LIST_REMOVE(op, nfso_list);
3432 		nfscl_freeopen(op, 0, false);
3433 	}
3434 	return (0);
3435 }
3436 
3437 /*
3438  * Return all delegations on this client.
3439  * (Must be called with client sleep lock.)
3440  */
3441 static void
3442 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p,
3443     struct nfscldeleghead *dhp)
3444 {
3445 	struct nfscldeleg *dp, *ndp;
3446 	struct ucred *cred;
3447 
3448 	cred = newnfs_getcred();
3449 	TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3450 		nfscl_cleandeleg(dp);
3451 		(void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3452 		if (dhp != NULL) {
3453 			nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
3454 			TAILQ_INSERT_HEAD(dhp, dp, nfsdl_list);
3455 		} else
3456 			nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
3457 	}
3458 	NFSFREECRED(cred);
3459 }
3460 
3461 /*
3462  * Return any delegation for this vp.
3463  */
3464 void
3465 nfscl_delegreturnvp(vnode_t vp, NFSPROC_T *p)
3466 {
3467 	struct nfsclclient *clp;
3468 	struct nfscldeleg *dp;
3469 	struct ucred *cred;
3470 	struct nfsnode *np;
3471 	struct nfsmount *nmp;
3472 
3473 	nmp = VFSTONFS(vp->v_mount);
3474 	NFSLOCKMNT(nmp);
3475 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
3476 		NFSUNLOCKMNT(nmp);
3477 		return;
3478 	}
3479 	NFSUNLOCKMNT(nmp);
3480 	np = VTONFS(vp);
3481 	cred = newnfs_getcred();
3482 	dp = NULL;
3483 	NFSLOCKCLSTATE();
3484 	clp = nmp->nm_clp;
3485 	if (clp != NULL)
3486 		dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
3487 		    np->n_fhp->nfh_len);
3488 	if (dp != NULL) {
3489 		nfscl_cleandeleg(dp);
3490 		nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
3491 		NFSUNLOCKCLSTATE();
3492 		newnfs_copycred(&dp->nfsdl_cred, cred);
3493 		nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3494 		free(dp, M_NFSCLDELEG);
3495 	} else
3496 		NFSUNLOCKCLSTATE();
3497 	NFSFREECRED(cred);
3498 }
3499 
3500 /*
3501  * Do a callback RPC.
3502  */
3503 void
3504 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3505 {
3506 	int clist, gotseq_ok, i, j, k, op, rcalls;
3507 	u_int32_t *tl;
3508 	struct nfsclclient *clp;
3509 	struct nfscldeleg *dp = NULL;
3510 	int numops, taglen = -1, error = 0, trunc __unused;
3511 	u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3512 	u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3513 	vnode_t vp = NULL;
3514 	struct nfsnode *np;
3515 	struct vattr va;
3516 	struct nfsfh *nfhp;
3517 	mount_t mp;
3518 	nfsattrbit_t attrbits, rattrbits;
3519 	nfsv4stateid_t stateid;
3520 	uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3521 	uint8_t sessionid[NFSX_V4SESSIONID];
3522 	struct mbuf *rep;
3523 	struct nfscllayout *lyp;
3524 	uint64_t filesid[2], len, off;
3525 	int changed, gotone, laytype, recalltype;
3526 	uint32_t iomode;
3527 	struct nfsclrecalllayout *recallp = NULL;
3528 	struct nfsclsession *tsep;
3529 
3530 	gotseq_ok = 0;
3531 	nfsrvd_rephead(nd);
3532 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3533 	taglen = fxdr_unsigned(int, *tl);
3534 	if (taglen < 0 || taglen > NFSV4_OPAQUELIMIT) {
3535 		error = EBADRPC;
3536 		taglen = -1;
3537 		goto nfsmout;
3538 	}
3539 	if (taglen <= NFSV4_SMALLSTR)
3540 		tagstr = tag;
3541 	else
3542 		tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3543 	error = nfsrv_mtostr(nd, tagstr, taglen);
3544 	if (error) {
3545 		if (taglen > NFSV4_SMALLSTR)
3546 			free(tagstr, M_TEMP);
3547 		taglen = -1;
3548 		goto nfsmout;
3549 	}
3550 	(void) nfsm_strtom(nd, tag, taglen);
3551 	if (taglen > NFSV4_SMALLSTR) {
3552 		free(tagstr, M_TEMP);
3553 	}
3554 	NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3555 	NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3556 	minorvers = fxdr_unsigned(u_int32_t, *tl++);
3557 	if (minorvers != NFSV4_MINORVERSION &&
3558 	    minorvers != NFSV41_MINORVERSION &&
3559 	    minorvers != NFSV42_MINORVERSION)
3560 		nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3561 	cbident = fxdr_unsigned(u_int32_t, *tl++);
3562 	if (nd->nd_repstat)
3563 		numops = 0;
3564 	else
3565 		numops = fxdr_unsigned(int, *tl);
3566 	/*
3567 	 * Loop around doing the sub ops.
3568 	 */
3569 	for (i = 0; i < numops; i++) {
3570 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3571 		NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3572 		*repp++ = *tl;
3573 		op = fxdr_unsigned(int, *tl);
3574 		nd->nd_procnum = op;
3575 		if (i == 0 && op != NFSV4OP_CBSEQUENCE && minorvers !=
3576 		    NFSV4_MINORVERSION) {
3577 		    nd->nd_repstat = NFSERR_OPNOTINSESS;
3578 		    *repp = nfscl_errmap(nd, minorvers);
3579 		    retops++;
3580 		    break;
3581 		}
3582 		if (op < NFSV4OP_CBGETATTR ||
3583 		   (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3584 		   (op > NFSV4OP_CBNOTIFYDEVID &&
3585 		    minorvers == NFSV41_MINORVERSION) ||
3586 		   (op > NFSV4OP_CBOFFLOAD &&
3587 		    minorvers == NFSV42_MINORVERSION)) {
3588 		    nd->nd_repstat = NFSERR_OPILLEGAL;
3589 		    *repp = nfscl_errmap(nd, minorvers);
3590 		    retops++;
3591 		    break;
3592 		}
3593 		if (op < NFSV42_CBNOPS)
3594 			nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3595 		switch (op) {
3596 		case NFSV4OP_CBGETATTR:
3597 			NFSCL_DEBUG(4, "cbgetattr\n");
3598 			mp = NULL;
3599 			vp = NULL;
3600 			error = nfsm_getfh(nd, &nfhp);
3601 			if (!error)
3602 				error = nfsrv_getattrbits(nd, &attrbits,
3603 				    NULL, NULL);
3604 			if (!error) {
3605 				mp = nfscl_getmnt(minorvers, sessionid, cbident,
3606 				    &clp);
3607 				if (mp == NULL)
3608 					error = NFSERR_SERVERFAULT;
3609 			}
3610 			if (!error) {
3611 				error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3612 				    nfhp->nfh_len, p, &np);
3613 				if (!error)
3614 					vp = NFSTOV(np);
3615 			}
3616 			if (!error) {
3617 				NFSZERO_ATTRBIT(&rattrbits);
3618 				NFSLOCKCLSTATE();
3619 				dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3620 				    nfhp->nfh_len);
3621 				if (dp != NULL) {
3622 					if (NFSISSET_ATTRBIT(&attrbits,
3623 					    NFSATTRBIT_SIZE)) {
3624 						if (vp != NULL)
3625 							va.va_size = np->n_size;
3626 						else
3627 							va.va_size =
3628 							    dp->nfsdl_size;
3629 						NFSSETBIT_ATTRBIT(&rattrbits,
3630 						    NFSATTRBIT_SIZE);
3631 					}
3632 					if (NFSISSET_ATTRBIT(&attrbits,
3633 					    NFSATTRBIT_CHANGE)) {
3634 						va.va_filerev =
3635 						    dp->nfsdl_change;
3636 						if (vp == NULL ||
3637 						    (np->n_flag & NDELEGMOD))
3638 							va.va_filerev++;
3639 						NFSSETBIT_ATTRBIT(&rattrbits,
3640 						    NFSATTRBIT_CHANGE);
3641 					}
3642 				} else
3643 					error = NFSERR_SERVERFAULT;
3644 				NFSUNLOCKCLSTATE();
3645 			}
3646 			if (vp != NULL)
3647 				vrele(vp);
3648 			if (mp != NULL)
3649 				vfs_unbusy(mp);
3650 			if (nfhp != NULL)
3651 				free(nfhp, M_NFSFH);
3652 			if (!error)
3653 				(void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3654 				    NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3655 				    (uint64_t)0, NULL);
3656 			break;
3657 		case NFSV4OP_CBRECALL:
3658 			NFSCL_DEBUG(4, "cbrecall\n");
3659 			NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3660 			    NFSX_UNSIGNED);
3661 			stateid.seqid = *tl++;
3662 			NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3663 			    NFSX_STATEIDOTHER);
3664 			tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3665 			trunc = fxdr_unsigned(int, *tl);
3666 			error = nfsm_getfh(nd, &nfhp);
3667 			if (!error) {
3668 				NFSLOCKCLSTATE();
3669 				if (minorvers == NFSV4_MINORVERSION)
3670 					clp = nfscl_getclnt(cbident);
3671 				else
3672 					clp = nfscl_getclntsess(sessionid);
3673 				if (clp != NULL) {
3674 					dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3675 					    nfhp->nfh_len);
3676 					if (dp != NULL && (dp->nfsdl_flags &
3677 					    NFSCLDL_DELEGRET) == 0) {
3678 						dp->nfsdl_flags |=
3679 						    NFSCLDL_RECALL;
3680 						wakeup((caddr_t)clp);
3681 					}
3682 				} else {
3683 					error = NFSERR_SERVERFAULT;
3684 				}
3685 				NFSUNLOCKCLSTATE();
3686 			}
3687 			if (nfhp != NULL)
3688 				free(nfhp, M_NFSFH);
3689 			break;
3690 		case NFSV4OP_CBLAYOUTRECALL:
3691 			NFSCL_DEBUG(4, "cblayrec\n");
3692 			nfhp = NULL;
3693 			NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3694 			laytype = fxdr_unsigned(int, *tl++);
3695 			iomode = fxdr_unsigned(uint32_t, *tl++);
3696 			if (newnfs_true == *tl++)
3697 				changed = 1;
3698 			else
3699 				changed = 0;
3700 			recalltype = fxdr_unsigned(int, *tl);
3701 			NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3702 			    laytype, iomode, changed, recalltype);
3703 			recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3704 			    M_WAITOK);
3705 			if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3706 			    laytype != NFSLAYOUT_FLEXFILE)
3707 				error = NFSERR_NOMATCHLAYOUT;
3708 			else if (recalltype == NFSLAYOUTRETURN_FILE) {
3709 				error = nfsm_getfh(nd, &nfhp);
3710 				NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3711 				if (error != 0)
3712 					goto nfsmout;
3713 				NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3714 				    NFSX_STATEID);
3715 				off = fxdr_hyper(tl); tl += 2;
3716 				len = fxdr_hyper(tl); tl += 2;
3717 				stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3718 				NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3719 				if (minorvers == NFSV4_MINORVERSION)
3720 					error = NFSERR_NOTSUPP;
3721 				NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3722 				    (uintmax_t)off, (uintmax_t)len,
3723 				    stateid.seqid, error);
3724 				if (error == 0) {
3725 					NFSLOCKCLSTATE();
3726 					clp = nfscl_getclntsess(sessionid);
3727 					NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3728 					if (clp != NULL) {
3729 						lyp = nfscl_findlayout(clp,
3730 						    nfhp->nfh_fh,
3731 						    nfhp->nfh_len);
3732 						NFSCL_DEBUG(4, "cblyp=%p\n",
3733 						    lyp);
3734 						if (lyp != NULL &&
3735 						    (lyp->nfsly_flags &
3736 						     (NFSLY_FILES |
3737 						      NFSLY_FLEXFILE)) != 0 &&
3738 						    !NFSBCMP(stateid.other,
3739 						    lyp->nfsly_stateid.other,
3740 						    NFSX_STATEIDOTHER)) {
3741 							error =
3742 							    nfscl_layoutrecall(
3743 							    recalltype,
3744 							    lyp, iomode, off,
3745 							    len, stateid.seqid,
3746 							    0, 0, NULL,
3747 							    recallp);
3748 							if (error == 0 &&
3749 							    stateid.seqid >
3750 							    lyp->nfsly_stateid.seqid)
3751 								lyp->nfsly_stateid.seqid =
3752 								    stateid.seqid;
3753 							recallp = NULL;
3754 							wakeup(clp);
3755 							NFSCL_DEBUG(4,
3756 							    "aft layrcal=%d "
3757 							    "layseqid=%d\n",
3758 							    error,
3759 							    lyp->nfsly_stateid.seqid);
3760 						} else
3761 							error =
3762 							  NFSERR_NOMATCHLAYOUT;
3763 					} else
3764 						error = NFSERR_NOMATCHLAYOUT;
3765 					NFSUNLOCKCLSTATE();
3766 				}
3767 				free(nfhp, M_NFSFH);
3768 			} else if (recalltype == NFSLAYOUTRETURN_FSID) {
3769 				NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3770 				filesid[0] = fxdr_hyper(tl); tl += 2;
3771 				filesid[1] = fxdr_hyper(tl); tl += 2;
3772 				gotone = 0;
3773 				NFSLOCKCLSTATE();
3774 				clp = nfscl_getclntsess(sessionid);
3775 				if (clp != NULL) {
3776 					TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3777 					    nfsly_list) {
3778 						if (lyp->nfsly_filesid[0] ==
3779 						    filesid[0] &&
3780 						    lyp->nfsly_filesid[1] ==
3781 						    filesid[1]) {
3782 							error =
3783 							    nfscl_layoutrecall(
3784 							    recalltype,
3785 							    lyp, iomode, 0,
3786 							    UINT64_MAX,
3787 							    lyp->nfsly_stateid.seqid,
3788 							    0, 0, NULL,
3789 							    recallp);
3790 							recallp = NULL;
3791 							gotone = 1;
3792 						}
3793 					}
3794 					if (gotone != 0)
3795 						wakeup(clp);
3796 					else
3797 						error = NFSERR_NOMATCHLAYOUT;
3798 				} else
3799 					error = NFSERR_NOMATCHLAYOUT;
3800 				NFSUNLOCKCLSTATE();
3801 			} else if (recalltype == NFSLAYOUTRETURN_ALL) {
3802 				gotone = 0;
3803 				NFSLOCKCLSTATE();
3804 				clp = nfscl_getclntsess(sessionid);
3805 				if (clp != NULL) {
3806 					TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3807 					    nfsly_list) {
3808 						error = nfscl_layoutrecall(
3809 						    recalltype, lyp, iomode, 0,
3810 						    UINT64_MAX,
3811 						    lyp->nfsly_stateid.seqid,
3812 						    0, 0, NULL, recallp);
3813 						recallp = NULL;
3814 						gotone = 1;
3815 					}
3816 					if (gotone != 0)
3817 						wakeup(clp);
3818 					else
3819 						error = NFSERR_NOMATCHLAYOUT;
3820 				} else
3821 					error = NFSERR_NOMATCHLAYOUT;
3822 				NFSUNLOCKCLSTATE();
3823 			} else
3824 				error = NFSERR_NOMATCHLAYOUT;
3825 			if (recallp != NULL) {
3826 				free(recallp, M_NFSLAYRECALL);
3827 				recallp = NULL;
3828 			}
3829 			break;
3830 		case NFSV4OP_CBSEQUENCE:
3831 			if (i != 0) {
3832 			    error = NFSERR_SEQUENCEPOS;
3833 			    break;
3834 			}
3835 			NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3836 			    5 * NFSX_UNSIGNED);
3837 			bcopy(tl, sessionid, NFSX_V4SESSIONID);
3838 			tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3839 			seqid = fxdr_unsigned(uint32_t, *tl++);
3840 			slotid = fxdr_unsigned(uint32_t, *tl++);
3841 			highslot = fxdr_unsigned(uint32_t, *tl++);
3842 			cachethis = *tl++;
3843 			/* Throw away the referring call stuff. */
3844 			clist = fxdr_unsigned(int, *tl);
3845 			for (j = 0; j < clist; j++) {
3846 				NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3847 				    NFSX_UNSIGNED);
3848 				tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3849 				rcalls = fxdr_unsigned(int, *tl);
3850 				for (k = 0; k < rcalls; k++) {
3851 					NFSM_DISSECT(tl, uint32_t *,
3852 					    2 * NFSX_UNSIGNED);
3853 				}
3854 			}
3855 			NFSLOCKCLSTATE();
3856 			clp = nfscl_getclntsess(sessionid);
3857 			if (clp == NULL)
3858 				error = NFSERR_SERVERFAULT;
3859 			if (error == 0) {
3860 				tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3861 				error = nfsv4_seqsession(seqid, slotid,
3862 				    highslot, tsep->nfsess_cbslots, &rep,
3863 				    tsep->nfsess_backslots);
3864 			}
3865 			NFSUNLOCKCLSTATE();
3866 			if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3867 				gotseq_ok = 1;
3868 				if (rep != NULL) {
3869 					/*
3870 					 * Handle a reply for a retried
3871 					 * callback.  The reply will be
3872 					 * re-inserted in the session cache
3873 					 * by the nfsv4_seqsess_cacherep() call
3874 					 * after out:
3875 					 */
3876 					KASSERT(error == NFSERR_REPLYFROMCACHE,
3877 					    ("cbsequence: non-NULL rep"));
3878 					NFSCL_DEBUG(4, "Got cbretry\n");
3879 					m_freem(nd->nd_mreq);
3880 					nd->nd_mreq = rep;
3881 					rep = NULL;
3882 					goto out;
3883 				}
3884 				NFSM_BUILD(tl, uint32_t *,
3885 				    NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3886 				bcopy(sessionid, tl, NFSX_V4SESSIONID);
3887 				tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3888 				*tl++ = txdr_unsigned(seqid);
3889 				*tl++ = txdr_unsigned(slotid);
3890 				*tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3891 				*tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3892 			}
3893 			break;
3894 		default:
3895 			if (i == 0 && minorvers != NFSV4_MINORVERSION)
3896 				error = NFSERR_OPNOTINSESS;
3897 			else {
3898 				NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3899 				error = NFSERR_NOTSUPP;
3900 			}
3901 			break;
3902 		}
3903 		if (error) {
3904 			if (error == EBADRPC || error == NFSERR_BADXDR) {
3905 				nd->nd_repstat = NFSERR_BADXDR;
3906 			} else {
3907 				nd->nd_repstat = error;
3908 			}
3909 			error = 0;
3910 		}
3911 		retops++;
3912 		if (nd->nd_repstat) {
3913 			*repp = nfscl_errmap(nd, minorvers);
3914 			break;
3915 		} else
3916 			*repp = 0;	/* NFS4_OK */
3917 	}
3918 nfsmout:
3919 	if (recallp != NULL)
3920 		free(recallp, M_NFSLAYRECALL);
3921 	if (error) {
3922 		if (error == EBADRPC || error == NFSERR_BADXDR)
3923 			nd->nd_repstat = NFSERR_BADXDR;
3924 		else
3925 			printf("nfsv4 comperr1=%d\n", error);
3926 	}
3927 	if (taglen == -1) {
3928 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3929 		*tl++ = 0;
3930 		*tl = 0;
3931 	} else {
3932 		*retopsp = txdr_unsigned(retops);
3933 	}
3934 	*nd->nd_errp = nfscl_errmap(nd, minorvers);
3935 out:
3936 	if (gotseq_ok != 0) {
3937 		rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3938 		NFSLOCKCLSTATE();
3939 		clp = nfscl_getclntsess(sessionid);
3940 		if (clp != NULL) {
3941 			tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3942 			nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3943 			    NFSERR_OK, &rep);
3944 			NFSUNLOCKCLSTATE();
3945 		} else {
3946 			NFSUNLOCKCLSTATE();
3947 			m_freem(rep);
3948 		}
3949 	}
3950 }
3951 
3952 /*
3953  * Generate the next cbident value. Basically just increment a static value
3954  * and then check that it isn't already in the list, if it has wrapped around.
3955  */
3956 static u_int32_t
3957 nfscl_nextcbident(void)
3958 {
3959 	struct nfsclclient *clp;
3960 	int matched;
3961 	static u_int32_t nextcbident = 0;
3962 	static int haswrapped = 0;
3963 
3964 	nextcbident++;
3965 	if (nextcbident == 0)
3966 		haswrapped = 1;
3967 	if (haswrapped) {
3968 		/*
3969 		 * Search the clientid list for one already using this cbident.
3970 		 */
3971 		do {
3972 			matched = 0;
3973 			NFSLOCKCLSTATE();
3974 			LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3975 				if (clp->nfsc_cbident == nextcbident) {
3976 					matched = 1;
3977 					break;
3978 				}
3979 			}
3980 			NFSUNLOCKCLSTATE();
3981 			if (matched == 1)
3982 				nextcbident++;
3983 		} while (matched);
3984 	}
3985 	return (nextcbident);
3986 }
3987 
3988 /*
3989  * Get the mount point related to a given cbident or session and busy it.
3990  */
3991 static mount_t
3992 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3993     struct nfsclclient **clpp)
3994 {
3995 	struct nfsclclient *clp;
3996 	mount_t mp;
3997 	int error;
3998 	struct nfsclsession *tsep;
3999 
4000 	*clpp = NULL;
4001 	NFSLOCKCLSTATE();
4002 	LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
4003 		tsep = nfsmnt_mdssession(clp->nfsc_nmp);
4004 		if (minorvers == NFSV4_MINORVERSION) {
4005 			if (clp->nfsc_cbident == cbident)
4006 				break;
4007 		} else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
4008 		    NFSX_V4SESSIONID))
4009 			break;
4010 	}
4011 	if (clp == NULL) {
4012 		NFSUNLOCKCLSTATE();
4013 		return (NULL);
4014 	}
4015 	mp = clp->nfsc_nmp->nm_mountp;
4016 	vfs_ref(mp);
4017 	NFSUNLOCKCLSTATE();
4018 	error = vfs_busy(mp, 0);
4019 	vfs_rel(mp);
4020 	if (error != 0)
4021 		return (NULL);
4022 	*clpp = clp;
4023 	return (mp);
4024 }
4025 
4026 /*
4027  * Get the clientid pointer related to a given cbident.
4028  */
4029 static struct nfsclclient *
4030 nfscl_getclnt(u_int32_t cbident)
4031 {
4032 	struct nfsclclient *clp;
4033 
4034 	LIST_FOREACH(clp, &nfsclhead, nfsc_list)
4035 		if (clp->nfsc_cbident == cbident)
4036 			break;
4037 	return (clp);
4038 }
4039 
4040 /*
4041  * Get the clientid pointer related to a given sessionid.
4042  */
4043 static struct nfsclclient *
4044 nfscl_getclntsess(uint8_t *sessionid)
4045 {
4046 	struct nfsclclient *clp;
4047 	struct nfsclsession *tsep;
4048 
4049 	LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
4050 		tsep = nfsmnt_mdssession(clp->nfsc_nmp);
4051 		if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
4052 		    NFSX_V4SESSIONID))
4053 			break;
4054 	}
4055 	return (clp);
4056 }
4057 
4058 /*
4059  * Search for a lock conflict locally on the client. A conflict occurs if
4060  * - not same owner and overlapping byte range and at least one of them is
4061  *   a write lock or this is an unlock.
4062  */
4063 static int
4064 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
4065     struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
4066     struct nfscllock **lopp)
4067 {
4068 	struct nfsclopen *op;
4069 	int ret;
4070 
4071 	if (dp != NULL) {
4072 		ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
4073 		if (ret)
4074 			return (ret);
4075 	}
4076 	LIST_FOREACH(op, NFSCLOPENHASH(clp, fhp, fhlen), nfso_hash) {
4077 		if (op->nfso_fhlen == fhlen &&
4078 		    !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
4079 			ret = nfscl_checkconflict(&op->nfso_lock, nlop,
4080 			    own, lopp);
4081 			if (ret)
4082 				return (ret);
4083 		}
4084 	}
4085 	return (0);
4086 }
4087 
4088 static int
4089 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
4090     u_int8_t *own, struct nfscllock **lopp)
4091 {
4092 	struct nfscllockowner *lp;
4093 	struct nfscllock *lop;
4094 
4095 	LIST_FOREACH(lp, lhp, nfsl_list) {
4096 		if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
4097 			LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
4098 				if (lop->nfslo_first >= nlop->nfslo_end)
4099 					break;
4100 				if (lop->nfslo_end <= nlop->nfslo_first)
4101 					continue;
4102 				if (lop->nfslo_type == F_WRLCK ||
4103 				    nlop->nfslo_type == F_WRLCK ||
4104 				    nlop->nfslo_type == F_UNLCK) {
4105 					if (lopp != NULL)
4106 						*lopp = lop;
4107 					return (NFSERR_DENIED);
4108 				}
4109 			}
4110 		}
4111 	}
4112 	return (0);
4113 }
4114 
4115 /*
4116  * Check for a local conflicting lock.
4117  */
4118 int
4119 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
4120     u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
4121 {
4122 	struct nfscllock *lop, nlck;
4123 	struct nfscldeleg *dp;
4124 	struct nfsnode *np;
4125 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
4126 	int error;
4127 
4128 	nlck.nfslo_type = fl->l_type;
4129 	nlck.nfslo_first = off;
4130 	if (len == NFS64BITSSET) {
4131 		nlck.nfslo_end = NFS64BITSSET;
4132 	} else {
4133 		nlck.nfslo_end = off + len;
4134 		if (nlck.nfslo_end <= nlck.nfslo_first)
4135 			return (NFSERR_INVAL);
4136 	}
4137 	np = VTONFS(vp);
4138 	nfscl_filllockowner(id, own, flags);
4139 	NFSLOCKCLSTATE();
4140 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4141 	error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
4142 	    &nlck, own, dp, &lop);
4143 	if (error != 0) {
4144 		fl->l_whence = SEEK_SET;
4145 		fl->l_start = lop->nfslo_first;
4146 		if (lop->nfslo_end == NFS64BITSSET)
4147 			fl->l_len = 0;
4148 		else
4149 			fl->l_len = lop->nfslo_end - lop->nfslo_first;
4150 		fl->l_pid = (pid_t)0;
4151 		fl->l_type = lop->nfslo_type;
4152 		error = -1;			/* no RPC required */
4153 	} else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
4154 	    fl->l_type == F_RDLCK)) {
4155 		/*
4156 		 * The delegation ensures that there isn't a conflicting
4157 		 * lock on the server, so return -1 to indicate an RPC
4158 		 * isn't required.
4159 		 */
4160 		fl->l_type = F_UNLCK;
4161 		error = -1;
4162 	}
4163 	NFSUNLOCKCLSTATE();
4164 	return (error);
4165 }
4166 
4167 /*
4168  * Handle Recall of a delegation.
4169  * The clp must be exclusive locked when this is called.
4170  */
4171 static int
4172 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
4173     struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
4174     int called_from_renewthread, vnode_t *vpp)
4175 {
4176 	struct nfsclowner *owp, *lowp, *nowp;
4177 	struct nfsclopen *op, *lop;
4178 	struct nfscllockowner *lp;
4179 	struct nfscllock *lckp;
4180 	struct nfsnode *np;
4181 	int error = 0, ret;
4182 
4183 	if (vp == NULL) {
4184 		KASSERT(vpp != NULL, ("nfscl_recalldeleg: vpp NULL"));
4185 		*vpp = NULL;
4186 		/*
4187 		 * First, get a vnode for the file. This is needed to do RPCs.
4188 		 */
4189 		ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
4190 		    dp->nfsdl_fhlen, p, &np);
4191 		if (ret) {
4192 			/*
4193 			 * File isn't open, so nothing to move over to the
4194 			 * server.
4195 			 */
4196 			return (0);
4197 		}
4198 		vp = NFSTOV(np);
4199 		*vpp = vp;
4200 	} else {
4201 		np = VTONFS(vp);
4202 	}
4203 	dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
4204 
4205 	/*
4206 	 * Ok, if it's a write delegation, flush data to the server, so
4207 	 * that close/open consistency is retained.
4208 	 */
4209 	ret = 0;
4210 	NFSLOCKNODE(np);
4211 	if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
4212 		np->n_flag |= NDELEGRECALL;
4213 		NFSUNLOCKNODE(np);
4214 		ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
4215 		NFSLOCKNODE(np);
4216 		np->n_flag &= ~NDELEGRECALL;
4217 	}
4218 	NFSINVALATTRCACHE(np);
4219 	NFSUNLOCKNODE(np);
4220 	if (ret == EIO && called_from_renewthread != 0) {
4221 		/*
4222 		 * If the flush failed with EIO for the renew thread,
4223 		 * return now, so that the dirty buffer will be flushed
4224 		 * later.
4225 		 */
4226 		return (ret);
4227 	}
4228 
4229 	/*
4230 	 * Now, for each openowner with opens issued locally, move them
4231 	 * over to state against the server.
4232 	 */
4233 	LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
4234 		lop = LIST_FIRST(&lowp->nfsow_open);
4235 		if (lop != NULL) {
4236 			if (LIST_NEXT(lop, nfso_list) != NULL)
4237 				panic("nfsdlg mult opens");
4238 			/*
4239 			 * Look for the same openowner against the server.
4240 			 */
4241 			LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
4242 				if (!NFSBCMP(lowp->nfsow_owner,
4243 				    owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
4244 					newnfs_copycred(&dp->nfsdl_cred, cred);
4245 					ret = nfscl_moveopen(vp, clp, nmp, lop,
4246 					    owp, dp, cred, p);
4247 					if (ret == NFSERR_STALECLIENTID ||
4248 					    ret == NFSERR_STALEDONTRECOVER ||
4249 					    ret == NFSERR_BADSESSION)
4250 						return (ret);
4251 					if (ret) {
4252 						nfscl_freeopen(lop, 1, true);
4253 						if (!error)
4254 							error = ret;
4255 					}
4256 					break;
4257 				}
4258 			}
4259 
4260 			/*
4261 			 * If no openowner found, create one and get an open
4262 			 * for it.
4263 			 */
4264 			if (owp == NULL) {
4265 				nowp = malloc(
4266 				    sizeof (struct nfsclowner), M_NFSCLOWNER,
4267 				    M_WAITOK);
4268 				nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4269 				    NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4270 				    dp->nfsdl_fhlen, NULL, NULL);
4271 				newnfs_copycred(&dp->nfsdl_cred, cred);
4272 				ret = nfscl_moveopen(vp, clp, nmp, lop,
4273 				    owp, dp, cred, p);
4274 				if (ret) {
4275 					nfscl_freeopenowner(owp, 0);
4276 					if (ret == NFSERR_STALECLIENTID ||
4277 					    ret == NFSERR_STALEDONTRECOVER ||
4278 					    ret == NFSERR_BADSESSION)
4279 						return (ret);
4280 					if (ret) {
4281 						nfscl_freeopen(lop, 1, true);
4282 						if (!error)
4283 							error = ret;
4284 					}
4285 				}
4286 			}
4287 		}
4288 	}
4289 
4290 	/*
4291 	 * Now, get byte range locks for any locks done locally.
4292 	 */
4293 	LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4294 		LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4295 			newnfs_copycred(&dp->nfsdl_cred, cred);
4296 			ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4297 			if (ret == NFSERR_STALESTATEID ||
4298 			    ret == NFSERR_STALEDONTRECOVER ||
4299 			    ret == NFSERR_STALECLIENTID ||
4300 			    ret == NFSERR_BADSESSION)
4301 				return (ret);
4302 			if (ret && !error)
4303 				error = ret;
4304 		}
4305 	}
4306 	return (error);
4307 }
4308 
4309 /*
4310  * Move a locally issued open over to an owner on the state list.
4311  * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4312  * returns with it unlocked.
4313  */
4314 static int
4315 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4316     struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4317     struct ucred *cred, NFSPROC_T *p)
4318 {
4319 	struct nfsclopen *op, *nop;
4320 	struct nfscldeleg *ndp;
4321 	struct nfsnode *np;
4322 	int error = 0, newone;
4323 
4324 	/*
4325 	 * First, look for an appropriate open, If found, just increment the
4326 	 * opencnt in it.
4327 	 */
4328 	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4329 		if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4330 		    op->nfso_fhlen == lop->nfso_fhlen &&
4331 		    !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4332 			op->nfso_opencnt += lop->nfso_opencnt;
4333 			nfscl_freeopen(lop, 1, true);
4334 			return (0);
4335 		}
4336 	}
4337 
4338 	/* No appropriate open, so we have to do one against the server. */
4339 	np = VTONFS(vp);
4340 	nop = malloc(sizeof (struct nfsclopen) +
4341 	    lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4342 	nop->nfso_hash.le_prev = NULL;
4343 	newone = 0;
4344 	nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4345 	    lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4346 	ndp = dp;
4347 	error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4348 	    lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4349 	    NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4350 	if (error) {
4351 		if (newone)
4352 			nfscl_freeopen(op, 0, true);
4353 	} else {
4354 		op->nfso_mode |= lop->nfso_mode;
4355 		op->nfso_opencnt += lop->nfso_opencnt;
4356 		nfscl_freeopen(lop, 1, true);
4357 	}
4358 	if (nop != NULL)
4359 		free(nop, M_NFSCLOPEN);
4360 	if (ndp != NULL) {
4361 		/*
4362 		 * What should I do with the returned delegation, since the
4363 		 * delegation is being recalled? For now, just printf and
4364 		 * through it away.
4365 		 */
4366 		printf("Moveopen returned deleg\n");
4367 		free(ndp, M_NFSCLDELEG);
4368 	}
4369 	return (error);
4370 }
4371 
4372 /*
4373  * Recall all delegations on this client.
4374  */
4375 static void
4376 nfscl_totalrecall(struct nfsclclient *clp)
4377 {
4378 	struct nfscldeleg *dp;
4379 
4380 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4381 		if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4382 			dp->nfsdl_flags |= NFSCLDL_RECALL;
4383 	}
4384 }
4385 
4386 /*
4387  * Relock byte ranges. Called for delegation recall and state expiry.
4388  */
4389 static int
4390 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4391     struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4392     NFSPROC_T *p)
4393 {
4394 	struct nfscllockowner *nlp;
4395 	struct nfsfh *nfhp;
4396 	struct nfsnode *np;
4397 	u_int64_t off, len;
4398 	int error, newone, donelocally;
4399 
4400 	if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) {
4401 		np = VTONFS(vp);
4402 		NFSLOCKNODE(np);
4403 		np->n_flag |= NMIGHTBELOCKED;
4404 		NFSUNLOCKNODE(np);
4405 	}
4406 
4407 	off = lop->nfslo_first;
4408 	len = lop->nfslo_end - lop->nfslo_first;
4409 	error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4410 	    clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4411 	    lp->nfsl_openowner, &nlp, &newone, &donelocally);
4412 	if (error || donelocally)
4413 		return (error);
4414 	nfhp = VTONFS(vp)->n_fhp;
4415 	error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4416 	    nfhp->nfh_len, nlp, newone, 0, off,
4417 	    len, lop->nfslo_type, cred, p);
4418 	if (error)
4419 		nfscl_freelockowner(nlp, 0);
4420 	return (error);
4421 }
4422 
4423 /*
4424  * Called to re-open a file. Basically get a vnode for the file handle
4425  * and then call nfsrpc_openrpc() to do the rest.
4426  */
4427 static int
4428 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4429     u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4430     struct ucred *cred, NFSPROC_T *p)
4431 {
4432 	struct nfsnode *np;
4433 	vnode_t vp;
4434 	int error;
4435 
4436 	error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4437 	if (error)
4438 		return (error);
4439 	vp = NFSTOV(np);
4440 	if (np->n_v4 != NULL) {
4441 		error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4442 		    np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4443 		    NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4444 		    cred, p);
4445 	} else {
4446 		error = EINVAL;
4447 	}
4448 	vrele(vp);
4449 	return (error);
4450 }
4451 
4452 /*
4453  * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4454  * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4455  * fail.
4456  */
4457 static int
4458 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4459     u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4460     u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4461     int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4462 {
4463 	int error;
4464 
4465 	do {
4466 		error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4467 		    mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4468 		    0, 0);
4469 		if (error == NFSERR_DELAY)
4470 			(void) nfs_catnap(PZERO, error, "nfstryop");
4471 	} while (error == NFSERR_DELAY);
4472 	if (error == EAUTH || error == EACCES) {
4473 		/* Try again using system credentials */
4474 		newnfs_setroot(cred);
4475 		do {
4476 		    error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4477 			newfhlen, mode, op, name, namelen, ndpp, reclaim,
4478 			delegtype, cred, p, 1, 0);
4479 		    if (error == NFSERR_DELAY)
4480 			(void) nfs_catnap(PZERO, error, "nfstryop");
4481 		} while (error == NFSERR_DELAY);
4482 	}
4483 	return (error);
4484 }
4485 
4486 /*
4487  * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4488  * NFSERR_DELAY. Also, retry with system credentials, if the provided
4489  * cred don't work.
4490  */
4491 static int
4492 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4493     int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4494     u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4495 {
4496 	struct nfsrv_descript nfsd, *nd = &nfsd;
4497 	int error;
4498 
4499 	do {
4500 		error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4501 		    reclaim, off, len, type, cred, p, 0);
4502 		if (!error && nd->nd_repstat == NFSERR_DELAY)
4503 			(void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4504 			    "nfstrylck");
4505 	} while (!error && nd->nd_repstat == NFSERR_DELAY);
4506 	if (!error)
4507 		error = nd->nd_repstat;
4508 	if (error == EAUTH || error == EACCES) {
4509 		/* Try again using root credentials */
4510 		newnfs_setroot(cred);
4511 		do {
4512 			error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4513 			    newone, reclaim, off, len, type, cred, p, 1);
4514 			if (!error && nd->nd_repstat == NFSERR_DELAY)
4515 				(void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4516 				    "nfstrylck");
4517 		} while (!error && nd->nd_repstat == NFSERR_DELAY);
4518 		if (!error)
4519 			error = nd->nd_repstat;
4520 	}
4521 	return (error);
4522 }
4523 
4524 /*
4525  * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4526  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4527  * credentials fail.
4528  */
4529 int
4530 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4531     struct nfsmount *nmp, NFSPROC_T *p)
4532 {
4533 	int error;
4534 
4535 	do {
4536 		error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4537 		if (error == NFSERR_DELAY)
4538 			(void) nfs_catnap(PZERO, error, "nfstrydp");
4539 	} while (error == NFSERR_DELAY);
4540 	if (error == EAUTH || error == EACCES) {
4541 		/* Try again using system credentials */
4542 		newnfs_setroot(cred);
4543 		do {
4544 			error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4545 			if (error == NFSERR_DELAY)
4546 				(void) nfs_catnap(PZERO, error, "nfstrydp");
4547 		} while (error == NFSERR_DELAY);
4548 	}
4549 	return (error);
4550 }
4551 
4552 /*
4553  * Try a close against the server. Just call nfsrpc_closerpc(),
4554  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4555  * credentials fail.
4556  */
4557 int
4558 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4559     struct nfsmount *nmp, NFSPROC_T *p, bool loop_on_delayed)
4560 {
4561 	struct nfsrv_descript nfsd, *nd = &nfsd;
4562 	int error;
4563 
4564 	do {
4565 		error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4566 		if (loop_on_delayed && error == NFSERR_DELAY)
4567 			(void) nfs_catnap(PZERO, error, "nfstrycl");
4568 	} while (loop_on_delayed && error == NFSERR_DELAY);
4569 	if (error == EAUTH || error == EACCES) {
4570 		/* Try again using system credentials */
4571 		newnfs_setroot(cred);
4572 		do {
4573 			error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4574 			if (loop_on_delayed && error == NFSERR_DELAY)
4575 				(void) nfs_catnap(PZERO, error, "nfstrycl");
4576 		} while (loop_on_delayed && error == NFSERR_DELAY);
4577 	}
4578 	return (error);
4579 }
4580 
4581 /*
4582  * Decide if a delegation on a file permits close without flushing writes
4583  * to the server. This might be a big performance win in some environments.
4584  * (Not useful until the client does caching on local stable storage.)
4585  */
4586 int
4587 nfscl_mustflush(vnode_t vp)
4588 {
4589 	struct nfsclclient *clp;
4590 	struct nfscldeleg *dp;
4591 	struct nfsnode *np;
4592 	struct nfsmount *nmp;
4593 
4594 	np = VTONFS(vp);
4595 	nmp = VFSTONFS(vp->v_mount);
4596 	if (!NFSHASNFSV4(nmp))
4597 		return (1);
4598 	NFSLOCKMNT(nmp);
4599 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4600 		NFSUNLOCKMNT(nmp);
4601 		return (1);
4602 	}
4603 	NFSUNLOCKMNT(nmp);
4604 	NFSLOCKCLSTATE();
4605 	clp = nfscl_findcl(nmp);
4606 	if (clp == NULL) {
4607 		NFSUNLOCKCLSTATE();
4608 		return (1);
4609 	}
4610 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4611 	if (dp != NULL && (dp->nfsdl_flags &
4612 	    (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4613 	     NFSCLDL_WRITE &&
4614 	    (dp->nfsdl_sizelimit >= np->n_size ||
4615 	     !NFSHASSTRICT3530(nmp))) {
4616 		NFSUNLOCKCLSTATE();
4617 		return (0);
4618 	}
4619 	NFSUNLOCKCLSTATE();
4620 	return (1);
4621 }
4622 
4623 /*
4624  * See if a (write) delegation exists for this file.
4625  */
4626 int
4627 nfscl_nodeleg(vnode_t vp, int writedeleg)
4628 {
4629 	struct nfsclclient *clp;
4630 	struct nfscldeleg *dp;
4631 	struct nfsnode *np;
4632 	struct nfsmount *nmp;
4633 
4634 	np = VTONFS(vp);
4635 	nmp = VFSTONFS(vp->v_mount);
4636 	if (!NFSHASNFSV4(nmp))
4637 		return (1);
4638 	NFSLOCKMNT(nmp);
4639 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4640 		NFSUNLOCKMNT(nmp);
4641 		return (1);
4642 	}
4643 	NFSUNLOCKMNT(nmp);
4644 	NFSLOCKCLSTATE();
4645 	clp = nfscl_findcl(nmp);
4646 	if (clp == NULL) {
4647 		NFSUNLOCKCLSTATE();
4648 		return (1);
4649 	}
4650 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4651 	if (dp != NULL &&
4652 	    (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4653 	    (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4654 	     NFSCLDL_WRITE)) {
4655 		NFSUNLOCKCLSTATE();
4656 		return (0);
4657 	}
4658 	NFSUNLOCKCLSTATE();
4659 	return (1);
4660 }
4661 
4662 /*
4663  * Look for an associated delegation that should be DelegReturned.
4664  */
4665 int
4666 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4667 {
4668 	struct nfsclclient *clp;
4669 	struct nfscldeleg *dp;
4670 	struct nfsclowner *owp;
4671 	struct nfscllockowner *lp;
4672 	struct nfsmount *nmp;
4673 	struct mount *mp;
4674 	struct ucred *cred;
4675 	struct nfsnode *np;
4676 	int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4677 
4678 	nmp = VFSTONFS(vp->v_mount);
4679 	if (NFSHASPNFS(nmp))
4680 		return (retcnt);
4681 	NFSLOCKMNT(nmp);
4682 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4683 		NFSUNLOCKMNT(nmp);
4684 		return (retcnt);
4685 	}
4686 	NFSUNLOCKMNT(nmp);
4687 	np = VTONFS(vp);
4688 	mp = nmp->nm_mountp;
4689 	NFSLOCKCLSTATE();
4690 	/*
4691 	 * Loop around waiting for:
4692 	 * - outstanding I/O operations on delegations to complete
4693 	 * - for a delegation on vp that has state, lock the client and
4694 	 *   do a recall
4695 	 * - return delegation with no state
4696 	 */
4697 	while (1) {
4698 		clp = nfscl_findcl(nmp);
4699 		if (clp == NULL) {
4700 			NFSUNLOCKCLSTATE();
4701 			return (retcnt);
4702 		}
4703 		dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4704 		    np->n_fhp->nfh_len);
4705 		if (dp != NULL) {
4706 		    /*
4707 		     * Wait for outstanding I/O ops to be done.
4708 		     */
4709 		    if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4710 			if (igotlock) {
4711 			    nfsv4_unlock(&clp->nfsc_lock, 0);
4712 			    igotlock = 0;
4713 			}
4714 			dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4715 			msleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO,
4716 			    "nfscld", hz);
4717 			if (NFSCL_FORCEDISM(mp)) {
4718 			    dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
4719 			    NFSUNLOCKCLSTATE();
4720 			    return (0);
4721 			}
4722 			continue;
4723 		    }
4724 		    needsrecall = 0;
4725 		    LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4726 			if (!LIST_EMPTY(&owp->nfsow_open)) {
4727 			    needsrecall = 1;
4728 			    break;
4729 			}
4730 		    }
4731 		    if (!needsrecall) {
4732 			LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4733 			    if (!LIST_EMPTY(&lp->nfsl_lock)) {
4734 				needsrecall = 1;
4735 				break;
4736 			    }
4737 			}
4738 		    }
4739 		    if (needsrecall && !triedrecall) {
4740 			dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4741 			islept = 0;
4742 			while (!igotlock) {
4743 			    igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4744 				&islept, NFSCLSTATEMUTEXPTR, mp);
4745 			    if (NFSCL_FORCEDISM(mp)) {
4746 				dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
4747 				if (igotlock)
4748 				    nfsv4_unlock(&clp->nfsc_lock, 0);
4749 				NFSUNLOCKCLSTATE();
4750 				return (0);
4751 			    }
4752 			    if (islept)
4753 				break;
4754 			}
4755 			if (islept)
4756 			    continue;
4757 			NFSUNLOCKCLSTATE();
4758 			cred = newnfs_getcred();
4759 			newnfs_copycred(&dp->nfsdl_cred, cred);
4760 			nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0, NULL);
4761 			NFSFREECRED(cred);
4762 			triedrecall = 1;
4763 			NFSLOCKCLSTATE();
4764 			nfsv4_unlock(&clp->nfsc_lock, 0);
4765 			igotlock = 0;
4766 			continue;
4767 		    }
4768 		    *stp = dp->nfsdl_stateid;
4769 		    retcnt = 1;
4770 		    nfscl_cleandeleg(dp);
4771 		    nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4772 		}
4773 		if (igotlock)
4774 		    nfsv4_unlock(&clp->nfsc_lock, 0);
4775 		NFSUNLOCKCLSTATE();
4776 		return (retcnt);
4777 	}
4778 }
4779 
4780 /*
4781  * Look for associated delegation(s) that should be DelegReturned.
4782  */
4783 int
4784 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4785     nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4786 {
4787 	struct nfsclclient *clp;
4788 	struct nfscldeleg *dp;
4789 	struct nfsclowner *owp;
4790 	struct nfscllockowner *lp;
4791 	struct nfsmount *nmp;
4792 	struct mount *mp;
4793 	struct ucred *cred;
4794 	struct nfsnode *np;
4795 	int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4796 
4797 	nmp = VFSTONFS(fvp->v_mount);
4798 	*gotfdp = 0;
4799 	*gottdp = 0;
4800 	if (NFSHASPNFS(nmp))
4801 		return (retcnt);
4802 	NFSLOCKMNT(nmp);
4803 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4804 		NFSUNLOCKMNT(nmp);
4805 		return (retcnt);
4806 	}
4807 	NFSUNLOCKMNT(nmp);
4808 	mp = nmp->nm_mountp;
4809 	NFSLOCKCLSTATE();
4810 	/*
4811 	 * Loop around waiting for:
4812 	 * - outstanding I/O operations on delegations to complete
4813 	 * - for a delegation on fvp that has state, lock the client and
4814 	 *   do a recall
4815 	 * - return delegation(s) with no state.
4816 	 */
4817 	while (1) {
4818 		clp = nfscl_findcl(nmp);
4819 		if (clp == NULL) {
4820 			NFSUNLOCKCLSTATE();
4821 			return (retcnt);
4822 		}
4823 		np = VTONFS(fvp);
4824 		dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4825 		    np->n_fhp->nfh_len);
4826 		if (dp != NULL && *gotfdp == 0) {
4827 		    /*
4828 		     * Wait for outstanding I/O ops to be done.
4829 		     */
4830 		    if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4831 			if (igotlock) {
4832 			    nfsv4_unlock(&clp->nfsc_lock, 0);
4833 			    igotlock = 0;
4834 			}
4835 			dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4836 			msleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO,
4837 			    "nfscld", hz);
4838 			if (NFSCL_FORCEDISM(mp)) {
4839 			    dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
4840 			    NFSUNLOCKCLSTATE();
4841 			    *gotfdp = 0;
4842 			    *gottdp = 0;
4843 			    return (0);
4844 			}
4845 			continue;
4846 		    }
4847 		    needsrecall = 0;
4848 		    LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4849 			if (!LIST_EMPTY(&owp->nfsow_open)) {
4850 			    needsrecall = 1;
4851 			    break;
4852 			}
4853 		    }
4854 		    if (!needsrecall) {
4855 			LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4856 			    if (!LIST_EMPTY(&lp->nfsl_lock)) {
4857 				needsrecall = 1;
4858 				break;
4859 			    }
4860 			}
4861 		    }
4862 		    if (needsrecall && !triedrecall) {
4863 			dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4864 			islept = 0;
4865 			while (!igotlock) {
4866 			    igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4867 				&islept, NFSCLSTATEMUTEXPTR, mp);
4868 			    if (NFSCL_FORCEDISM(mp)) {
4869 				dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
4870 				if (igotlock)
4871 				    nfsv4_unlock(&clp->nfsc_lock, 0);
4872 				NFSUNLOCKCLSTATE();
4873 				*gotfdp = 0;
4874 				*gottdp = 0;
4875 				return (0);
4876 			    }
4877 			    if (islept)
4878 				break;
4879 			}
4880 			if (islept)
4881 			    continue;
4882 			NFSUNLOCKCLSTATE();
4883 			cred = newnfs_getcred();
4884 			newnfs_copycred(&dp->nfsdl_cred, cred);
4885 			nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0, NULL);
4886 			NFSFREECRED(cred);
4887 			triedrecall = 1;
4888 			NFSLOCKCLSTATE();
4889 			nfsv4_unlock(&clp->nfsc_lock, 0);
4890 			igotlock = 0;
4891 			continue;
4892 		    }
4893 		    *fstp = dp->nfsdl_stateid;
4894 		    retcnt++;
4895 		    *gotfdp = 1;
4896 		    nfscl_cleandeleg(dp);
4897 		    nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4898 		}
4899 		if (igotlock) {
4900 		    nfsv4_unlock(&clp->nfsc_lock, 0);
4901 		    igotlock = 0;
4902 		}
4903 		if (tvp != NULL) {
4904 		    np = VTONFS(tvp);
4905 		    dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4906 			np->n_fhp->nfh_len);
4907 		    if (dp != NULL && *gottdp == 0) {
4908 			/*
4909 			 * Wait for outstanding I/O ops to be done.
4910 			 */
4911 			if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4912 			    dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4913 			    msleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO,
4914 				"nfscld", hz);
4915 			    if (NFSCL_FORCEDISM(mp)) {
4916 				NFSUNLOCKCLSTATE();
4917 				*gotfdp = 0;
4918 				*gottdp = 0;
4919 				return (0);
4920 			    }
4921 			    continue;
4922 			}
4923 			LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4924 			    if (!LIST_EMPTY(&owp->nfsow_open)) {
4925 				NFSUNLOCKCLSTATE();
4926 				return (retcnt);
4927 			    }
4928 			}
4929 			LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4930 			    if (!LIST_EMPTY(&lp->nfsl_lock)) {
4931 				NFSUNLOCKCLSTATE();
4932 				return (retcnt);
4933 			    }
4934 			}
4935 			*tstp = dp->nfsdl_stateid;
4936 			retcnt++;
4937 			*gottdp = 1;
4938 			nfscl_cleandeleg(dp);
4939 			nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4940 		    }
4941 		}
4942 		NFSUNLOCKCLSTATE();
4943 		return (retcnt);
4944 	}
4945 }
4946 
4947 /*
4948  * Get a reference on the clientid associated with the mount point.
4949  * Return 1 if success, 0 otherwise.
4950  */
4951 int
4952 nfscl_getref(struct nfsmount *nmp)
4953 {
4954 	struct nfsclclient *clp;
4955 	int ret;
4956 
4957 	NFSLOCKCLSTATE();
4958 	clp = nfscl_findcl(nmp);
4959 	if (clp == NULL) {
4960 		NFSUNLOCKCLSTATE();
4961 		return (0);
4962 	}
4963 	nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, nmp->nm_mountp);
4964 	ret = 1;
4965 	if (NFSCL_FORCEDISM(nmp->nm_mountp))
4966 		ret = 0;
4967 	NFSUNLOCKCLSTATE();
4968 	return (ret);
4969 }
4970 
4971 /*
4972  * Release a reference on a clientid acquired with the above call.
4973  */
4974 void
4975 nfscl_relref(struct nfsmount *nmp)
4976 {
4977 	struct nfsclclient *clp;
4978 
4979 	NFSLOCKCLSTATE();
4980 	clp = nfscl_findcl(nmp);
4981 	if (clp == NULL) {
4982 		NFSUNLOCKCLSTATE();
4983 		return;
4984 	}
4985 	nfsv4_relref(&clp->nfsc_lock);
4986 	NFSUNLOCKCLSTATE();
4987 }
4988 
4989 /*
4990  * Save the size attribute in the delegation, since the nfsnode
4991  * is going away.
4992  */
4993 void
4994 nfscl_reclaimnode(vnode_t vp)
4995 {
4996 	struct nfsclclient *clp;
4997 	struct nfscldeleg *dp;
4998 	struct nfsnode *np = VTONFS(vp);
4999 	struct nfsmount *nmp;
5000 
5001 	nmp = VFSTONFS(vp->v_mount);
5002 	if (!NFSHASNFSV4(nmp))
5003 		return;
5004 	NFSLOCKCLSTATE();
5005 	clp = nfscl_findcl(nmp);
5006 	if (clp == NULL) {
5007 		NFSUNLOCKCLSTATE();
5008 		return;
5009 	}
5010 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5011 	if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
5012 		dp->nfsdl_size = np->n_size;
5013 	NFSUNLOCKCLSTATE();
5014 }
5015 
5016 /*
5017  * Get the saved size attribute in the delegation, since it is a
5018  * newly allocated nfsnode.
5019  */
5020 void
5021 nfscl_newnode(vnode_t vp)
5022 {
5023 	struct nfsclclient *clp;
5024 	struct nfscldeleg *dp;
5025 	struct nfsnode *np = VTONFS(vp);
5026 	struct nfsmount *nmp;
5027 
5028 	nmp = VFSTONFS(vp->v_mount);
5029 	if (!NFSHASNFSV4(nmp))
5030 		return;
5031 	NFSLOCKCLSTATE();
5032 	clp = nfscl_findcl(nmp);
5033 	if (clp == NULL) {
5034 		NFSUNLOCKCLSTATE();
5035 		return;
5036 	}
5037 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5038 	if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
5039 		np->n_size = dp->nfsdl_size;
5040 	NFSUNLOCKCLSTATE();
5041 }
5042 
5043 /*
5044  * If there is a valid write delegation for this file, set the modtime
5045  * to the local clock time.
5046  */
5047 void
5048 nfscl_delegmodtime(vnode_t vp)
5049 {
5050 	struct nfsclclient *clp;
5051 	struct nfscldeleg *dp;
5052 	struct nfsnode *np = VTONFS(vp);
5053 	struct nfsmount *nmp;
5054 
5055 	nmp = VFSTONFS(vp->v_mount);
5056 	if (!NFSHASNFSV4(nmp))
5057 		return;
5058 	NFSLOCKMNT(nmp);
5059 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
5060 		NFSUNLOCKMNT(nmp);
5061 		return;
5062 	}
5063 	NFSUNLOCKMNT(nmp);
5064 	NFSLOCKCLSTATE();
5065 	clp = nfscl_findcl(nmp);
5066 	if (clp == NULL) {
5067 		NFSUNLOCKCLSTATE();
5068 		return;
5069 	}
5070 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5071 	if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
5072 		nanotime(&dp->nfsdl_modtime);
5073 		dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
5074 	}
5075 	NFSUNLOCKCLSTATE();
5076 }
5077 
5078 /*
5079  * If there is a valid write delegation for this file with a modtime set,
5080  * put that modtime in mtime.
5081  */
5082 void
5083 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
5084 {
5085 	struct nfsclclient *clp;
5086 	struct nfscldeleg *dp;
5087 	struct nfsnode *np = VTONFS(vp);
5088 	struct nfsmount *nmp;
5089 
5090 	nmp = VFSTONFS(vp->v_mount);
5091 	if (!NFSHASNFSV4(nmp))
5092 		return;
5093 	NFSLOCKMNT(nmp);
5094 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
5095 		NFSUNLOCKMNT(nmp);
5096 		return;
5097 	}
5098 	NFSUNLOCKMNT(nmp);
5099 	NFSLOCKCLSTATE();
5100 	clp = nfscl_findcl(nmp);
5101 	if (clp == NULL) {
5102 		NFSUNLOCKCLSTATE();
5103 		return;
5104 	}
5105 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5106 	if (dp != NULL &&
5107 	    (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
5108 	    (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
5109 		*mtime = dp->nfsdl_modtime;
5110 	NFSUNLOCKCLSTATE();
5111 }
5112 
5113 static int
5114 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
5115 {
5116 	short *defaulterrp, *errp;
5117 
5118 	if (!nd->nd_repstat)
5119 		return (0);
5120 	if (nd->nd_procnum == NFSPROC_NOOP)
5121 		return (txdr_unsigned(nd->nd_repstat & 0xffff));
5122 	if (nd->nd_repstat == EBADRPC)
5123 		return (txdr_unsigned(NFSERR_BADXDR));
5124 	if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
5125 	    nd->nd_repstat == NFSERR_OPILLEGAL)
5126 		return (txdr_unsigned(nd->nd_repstat));
5127 	if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
5128 	    minorvers > NFSV4_MINORVERSION) {
5129 		/* NFSv4.n error. */
5130 		return (txdr_unsigned(nd->nd_repstat));
5131 	}
5132 	if (nd->nd_procnum < NFSV4OP_CBNOPS)
5133 		errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
5134 	else
5135 		return (txdr_unsigned(nd->nd_repstat));
5136 	while (*++errp)
5137 		if (*errp == (short)nd->nd_repstat)
5138 			return (txdr_unsigned(nd->nd_repstat));
5139 	return (txdr_unsigned(*defaulterrp));
5140 }
5141 
5142 /*
5143  * Called to find/add a layout to a client.
5144  * This function returns the layout with a refcnt (shared lock) upon
5145  * success (returns 0) or with no lock/refcnt on the layout when an
5146  * error is returned.
5147  * If a layout is passed in via lypp, it is locked (exclusively locked).
5148  */
5149 int
5150 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
5151     nfsv4stateid_t *stateidp, int layouttype, int retonclose,
5152     struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
5153     struct ucred *cred, NFSPROC_T *p)
5154 {
5155 	struct nfsclclient *clp;
5156 	struct nfscllayout *lyp, *tlyp;
5157 	struct nfsclflayout *flp;
5158 	struct nfsnode *np = VTONFS(vp);
5159 	mount_t mp;
5160 	int layout_passed_in;
5161 
5162 	mp = nmp->nm_mountp;
5163 	layout_passed_in = 1;
5164 	tlyp = NULL;
5165 	lyp = *lypp;
5166 	if (lyp == NULL) {
5167 		layout_passed_in = 0;
5168 		tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
5169 		    M_WAITOK | M_ZERO);
5170 	}
5171 
5172 	NFSLOCKCLSTATE();
5173 	clp = nmp->nm_clp;
5174 	if (clp == NULL) {
5175 		if (layout_passed_in != 0)
5176 			nfsv4_unlock(&lyp->nfsly_lock, 0);
5177 		NFSUNLOCKCLSTATE();
5178 		if (tlyp != NULL)
5179 			free(tlyp, M_NFSLAYOUT);
5180 		return (EPERM);
5181 	}
5182 	if (lyp == NULL) {
5183 		/*
5184 		 * Although no lyp was passed in, another thread might have
5185 		 * allocated one. If one is found, just increment it's ref
5186 		 * count and return it.
5187 		 */
5188 		lyp = nfscl_findlayout(clp, fhp, fhlen);
5189 		if (lyp == NULL) {
5190 			lyp = tlyp;
5191 			tlyp = NULL;
5192 			lyp->nfsly_stateid.seqid = stateidp->seqid;
5193 			lyp->nfsly_stateid.other[0] = stateidp->other[0];
5194 			lyp->nfsly_stateid.other[1] = stateidp->other[1];
5195 			lyp->nfsly_stateid.other[2] = stateidp->other[2];
5196 			lyp->nfsly_lastbyte = 0;
5197 			LIST_INIT(&lyp->nfsly_flayread);
5198 			LIST_INIT(&lyp->nfsly_flayrw);
5199 			LIST_INIT(&lyp->nfsly_recall);
5200 			lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
5201 			lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
5202 			lyp->nfsly_clp = clp;
5203 			if (layouttype == NFSLAYOUT_FLEXFILE)
5204 				lyp->nfsly_flags = NFSLY_FLEXFILE;
5205 			else
5206 				lyp->nfsly_flags = NFSLY_FILES;
5207 			if (retonclose != 0)
5208 				lyp->nfsly_flags |= NFSLY_RETONCLOSE;
5209 			lyp->nfsly_fhlen = fhlen;
5210 			NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
5211 			TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5212 			LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
5213 			    nfsly_hash);
5214 			lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5215 			nfscl_layoutcnt++;
5216 			nfsstatsv1.cllayouts++;
5217 		} else {
5218 			if (retonclose != 0)
5219 				lyp->nfsly_flags |= NFSLY_RETONCLOSE;
5220 			if (stateidp->seqid > lyp->nfsly_stateid.seqid)
5221 				lyp->nfsly_stateid.seqid = stateidp->seqid;
5222 			TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5223 			TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5224 			lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5225 		}
5226 		nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5227 		if (NFSCL_FORCEDISM(mp)) {
5228 			NFSUNLOCKCLSTATE();
5229 			if (tlyp != NULL)
5230 				free(tlyp, M_NFSLAYOUT);
5231 			return (EPERM);
5232 		}
5233 		*lypp = lyp;
5234 	} else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
5235 		lyp->nfsly_stateid.seqid = stateidp->seqid;
5236 
5237 	/* Merge the new list of File Layouts into the list. */
5238 	flp = LIST_FIRST(fhlp);
5239 	if (flp != NULL) {
5240 		if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
5241 			nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
5242 		else
5243 			nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
5244 	}
5245 	if (layout_passed_in != 0)
5246 		nfsv4_unlock(&lyp->nfsly_lock, 1);
5247 	NFSUNLOCKCLSTATE();
5248 	if (tlyp != NULL)
5249 		free(tlyp, M_NFSLAYOUT);
5250 	return (0);
5251 }
5252 
5253 /*
5254  * Search for a layout by MDS file handle.
5255  * If one is found, it is returned with a refcnt (shared lock) iff
5256  * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
5257  * returned NULL.
5258  */
5259 struct nfscllayout *
5260 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
5261     uint64_t off, uint32_t rwaccess, struct nfsclflayout **retflpp,
5262     int *recalledp)
5263 {
5264 	struct nfscllayout *lyp;
5265 	mount_t mp;
5266 	int error, igotlock;
5267 
5268 	mp = clp->nfsc_nmp->nm_mountp;
5269 	*recalledp = 0;
5270 	*retflpp = NULL;
5271 	NFSLOCKCLSTATE();
5272 	lyp = nfscl_findlayout(clp, fhp, fhlen);
5273 	if (lyp != NULL) {
5274 		if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5275 			TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5276 			TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5277 			lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5278 			error = nfscl_findlayoutforio(lyp, off, rwaccess,
5279 			    retflpp);
5280 			if (error == 0)
5281 				nfsv4_getref(&lyp->nfsly_lock, NULL,
5282 				    NFSCLSTATEMUTEXPTR, mp);
5283 			else {
5284 				do {
5285 					igotlock = nfsv4_lock(&lyp->nfsly_lock,
5286 					    1, NULL, NFSCLSTATEMUTEXPTR, mp);
5287 				} while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
5288 				*retflpp = NULL;
5289 			}
5290 			if (NFSCL_FORCEDISM(mp)) {
5291 				lyp = NULL;
5292 				*recalledp = 1;
5293 			}
5294 		} else {
5295 			lyp = NULL;
5296 			*recalledp = 1;
5297 		}
5298 	}
5299 	NFSUNLOCKCLSTATE();
5300 	return (lyp);
5301 }
5302 
5303 /*
5304  * Search for a layout by MDS file handle. If one is found, mark in to be
5305  * recalled, if it already marked "return on close".
5306  */
5307 static void
5308 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
5309     int fhlen, struct nfsclrecalllayout **recallpp, struct nfscllayout **lypp)
5310 {
5311 	struct nfscllayout *lyp;
5312 	uint32_t iomode;
5313 
5314 	*lypp = NULL;
5315 	if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vp->v_mount)) ||
5316 	    nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
5317 	    (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
5318 		return;
5319 	lyp = nfscl_findlayout(clp, fhp, fhlen);
5320 	if (lyp != NULL && (lyp->nfsly_flags & NFSLY_RETONCLOSE) != 0) {
5321 		if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5322 			iomode = 0;
5323 			if (!LIST_EMPTY(&lyp->nfsly_flayread))
5324 				iomode |= NFSLAYOUTIOMODE_READ;
5325 			if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5326 				iomode |= NFSLAYOUTIOMODE_RW;
5327 			nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5328 			    0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
5329 			    *recallpp);
5330 			NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
5331 			*recallpp = NULL;
5332 		}
5333 
5334 		/* Now, wake up renew thread to do LayoutReturn. */
5335 		wakeup(clp);
5336 		*lypp = lyp;
5337 	}
5338 }
5339 
5340 /*
5341  * Mark the layout to be recalled and with an error.
5342  * Also, disable the dsp from further use.
5343  */
5344 void
5345 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
5346     struct nfscllayout *lyp, struct nfsclds *dsp)
5347 {
5348 	struct nfsclrecalllayout *recallp;
5349 	uint32_t iomode;
5350 
5351 	printf("DS being disabled, error=%d\n", stat);
5352 	/* Set up the return of the layout. */
5353 	recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
5354 	iomode = 0;
5355 	NFSLOCKCLSTATE();
5356 	if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5357 		if (!LIST_EMPTY(&lyp->nfsly_flayread))
5358 			iomode |= NFSLAYOUTIOMODE_READ;
5359 		if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5360 			iomode |= NFSLAYOUTIOMODE_RW;
5361 		(void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5362 		    0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5363 		    dp->nfsdi_deviceid, recallp);
5364 		NFSUNLOCKCLSTATE();
5365 		NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5366 	} else {
5367 		NFSUNLOCKCLSTATE();
5368 		free(recallp, M_NFSLAYRECALL);
5369 	}
5370 
5371 	/* And shut the TCP connection down. */
5372 	nfscl_cancelreqs(dsp);
5373 }
5374 
5375 /*
5376  * Cancel all RPCs for this "dsp" by closing the connection.
5377  * Also, mark the session as defunct.
5378  * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5379  * cannot be shut down.
5380  */
5381 void
5382 nfscl_cancelreqs(struct nfsclds *dsp)
5383 {
5384 	struct __rpc_client *cl;
5385 	static int non_event;
5386 
5387 	NFSLOCKDS(dsp);
5388 	if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5389 	    dsp->nfsclds_sockp != NULL &&
5390 	    dsp->nfsclds_sockp->nr_client != NULL) {
5391 		dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5392 		cl = dsp->nfsclds_sockp->nr_client;
5393 		dsp->nfsclds_sess.nfsess_defunct = 1;
5394 		NFSUNLOCKDS(dsp);
5395 		CLNT_CLOSE(cl);
5396 		/*
5397 		 * This 1sec sleep is done to reduce the number of reconnect
5398 		 * attempts made on the DS while it has failed.
5399 		 */
5400 		tsleep(&non_event, PVFS, "ndscls", hz);
5401 		return;
5402 	}
5403 	NFSUNLOCKDS(dsp);
5404 }
5405 
5406 /*
5407  * Dereference a layout.
5408  */
5409 void
5410 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5411 {
5412 
5413 	NFSLOCKCLSTATE();
5414 	if (exclocked != 0)
5415 		nfsv4_unlock(&lyp->nfsly_lock, 0);
5416 	else
5417 		nfsv4_relref(&lyp->nfsly_lock);
5418 	NFSUNLOCKCLSTATE();
5419 }
5420 
5421 /*
5422  * Search for a devinfo by deviceid. If one is found, return it after
5423  * acquiring a reference count on it.
5424  */
5425 struct nfscldevinfo *
5426 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5427     struct nfscldevinfo *dip)
5428 {
5429 
5430 	NFSLOCKCLSTATE();
5431 	if (dip == NULL)
5432 		dip = nfscl_finddevinfo(clp, deviceid);
5433 	if (dip != NULL)
5434 		dip->nfsdi_refcnt++;
5435 	NFSUNLOCKCLSTATE();
5436 	return (dip);
5437 }
5438 
5439 /*
5440  * Dereference a devinfo structure.
5441  */
5442 static void
5443 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5444 {
5445 
5446 	dip->nfsdi_refcnt--;
5447 	if (dip->nfsdi_refcnt == 0)
5448 		wakeup(&dip->nfsdi_refcnt);
5449 }
5450 
5451 /*
5452  * Dereference a devinfo structure.
5453  */
5454 void
5455 nfscl_reldevinfo(struct nfscldevinfo *dip)
5456 {
5457 
5458 	NFSLOCKCLSTATE();
5459 	nfscl_reldevinfo_locked(dip);
5460 	NFSUNLOCKCLSTATE();
5461 }
5462 
5463 /*
5464  * Find a layout for this file handle. Return NULL upon failure.
5465  */
5466 static struct nfscllayout *
5467 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5468 {
5469 	struct nfscllayout *lyp;
5470 
5471 	LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5472 		if (lyp->nfsly_fhlen == fhlen &&
5473 		    !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5474 			break;
5475 	return (lyp);
5476 }
5477 
5478 /*
5479  * Find a devinfo for this deviceid. Return NULL upon failure.
5480  */
5481 static struct nfscldevinfo *
5482 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5483 {
5484 	struct nfscldevinfo *dip;
5485 
5486 	LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5487 		if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5488 		    == 0)
5489 			break;
5490 	return (dip);
5491 }
5492 
5493 /*
5494  * Merge the new file layout list into the main one, maintaining it in
5495  * increasing offset order.
5496  */
5497 static void
5498 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5499     struct nfsclflayouthead *newfhlp)
5500 {
5501 	struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5502 
5503 	flp = LIST_FIRST(fhlp);
5504 	prevflp = NULL;
5505 	LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5506 		while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5507 			prevflp = flp;
5508 			flp = LIST_NEXT(flp, nfsfl_list);
5509 		}
5510 		if (prevflp == NULL)
5511 			LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5512 		else
5513 			LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5514 		prevflp = nflp;
5515 	}
5516 }
5517 
5518 /*
5519  * Add this nfscldevinfo to the client, if it doesn't already exist.
5520  * This function consumes the structure pointed at by dip, if not NULL.
5521  */
5522 int
5523 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5524     struct nfsclflayout *flp)
5525 {
5526 	struct nfsclclient *clp;
5527 	struct nfscldevinfo *tdip;
5528 	uint8_t *dev;
5529 
5530 	NFSLOCKCLSTATE();
5531 	clp = nmp->nm_clp;
5532 	if (clp == NULL) {
5533 		NFSUNLOCKCLSTATE();
5534 		if (dip != NULL)
5535 			free(dip, M_NFSDEVINFO);
5536 		return (ENODEV);
5537 	}
5538 	if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5539 		dev = flp->nfsfl_dev;
5540 	else
5541 		dev = flp->nfsfl_ffm[ind].dev;
5542 	tdip = nfscl_finddevinfo(clp, dev);
5543 	if (tdip != NULL) {
5544 		tdip->nfsdi_layoutrefs++;
5545 		if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5546 			flp->nfsfl_devp = tdip;
5547 		else
5548 			flp->nfsfl_ffm[ind].devp = tdip;
5549 		nfscl_reldevinfo_locked(tdip);
5550 		NFSUNLOCKCLSTATE();
5551 		if (dip != NULL)
5552 			free(dip, M_NFSDEVINFO);
5553 		return (0);
5554 	}
5555 	if (dip != NULL) {
5556 		LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5557 		dip->nfsdi_layoutrefs = 1;
5558 		if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5559 			flp->nfsfl_devp = dip;
5560 		else
5561 			flp->nfsfl_ffm[ind].devp = dip;
5562 	}
5563 	NFSUNLOCKCLSTATE();
5564 	if (dip == NULL)
5565 		return (ENODEV);
5566 	return (0);
5567 }
5568 
5569 /*
5570  * Free up a layout structure and associated file layout structure(s).
5571  */
5572 void
5573 nfscl_freelayout(struct nfscllayout *layp)
5574 {
5575 	struct nfsclflayout *flp, *nflp;
5576 	struct nfsclrecalllayout *rp, *nrp;
5577 
5578 	LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5579 		LIST_REMOVE(flp, nfsfl_list);
5580 		nfscl_freeflayout(flp);
5581 	}
5582 	LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5583 		LIST_REMOVE(flp, nfsfl_list);
5584 		nfscl_freeflayout(flp);
5585 	}
5586 	LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5587 		LIST_REMOVE(rp, nfsrecly_list);
5588 		free(rp, M_NFSLAYRECALL);
5589 	}
5590 	nfscl_layoutcnt--;
5591 	nfsstatsv1.cllayouts--;
5592 	free(layp, M_NFSLAYOUT);
5593 }
5594 
5595 /*
5596  * Free up a file layout structure.
5597  */
5598 void
5599 nfscl_freeflayout(struct nfsclflayout *flp)
5600 {
5601 	int i, j;
5602 
5603 	if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5604 		for (i = 0; i < flp->nfsfl_fhcnt; i++)
5605 			free(flp->nfsfl_fh[i], M_NFSFH);
5606 		if (flp->nfsfl_devp != NULL)
5607 			flp->nfsfl_devp->nfsdi_layoutrefs--;
5608 	}
5609 	if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5610 		for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5611 			for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5612 				free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5613 			if (flp->nfsfl_ffm[i].devp != NULL)
5614 				flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5615 		}
5616 	free(flp, M_NFSFLAYOUT);
5617 }
5618 
5619 /*
5620  * Free up a file layout devinfo structure.
5621  */
5622 void
5623 nfscl_freedevinfo(struct nfscldevinfo *dip)
5624 {
5625 
5626 	free(dip, M_NFSDEVINFO);
5627 }
5628 
5629 /*
5630  * Mark any layouts that match as recalled.
5631  */
5632 static int
5633 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5634     uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5635     char *devid, struct nfsclrecalllayout *recallp)
5636 {
5637 	struct nfsclrecalllayout *rp, *orp;
5638 
5639 	recallp->nfsrecly_recalltype = recalltype;
5640 	recallp->nfsrecly_iomode = iomode;
5641 	recallp->nfsrecly_stateseqid = stateseqid;
5642 	recallp->nfsrecly_off = off;
5643 	recallp->nfsrecly_len = len;
5644 	recallp->nfsrecly_stat = stat;
5645 	recallp->nfsrecly_op = op;
5646 	if (devid != NULL)
5647 		NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5648 	/*
5649 	 * Order the list as file returns first, followed by fsid and any
5650 	 * returns, both in increasing stateseqid order.
5651 	 * Note that the seqids wrap around, so 1 is after 0xffffffff.
5652 	 * (I'm not sure this is correct because I find RFC5661 confusing
5653 	 *  on this, but hopefully it will work ok.)
5654 	 */
5655 	orp = NULL;
5656 	LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5657 		orp = rp;
5658 		if ((recalltype == NFSLAYOUTRETURN_FILE &&
5659 		     (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5660 		      nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5661 		    (recalltype != NFSLAYOUTRETURN_FILE &&
5662 		     rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5663 		     nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5664 			LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5665 			break;
5666 		}
5667 
5668 		/*
5669 		 * Put any error return on all the file returns that will
5670 		 * preceed this one.
5671 		 */
5672 		if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5673 		   stat != 0 && rp->nfsrecly_stat == 0) {
5674 			rp->nfsrecly_stat = stat;
5675 			rp->nfsrecly_op = op;
5676 			if (devid != NULL)
5677 				NFSBCOPY(devid, rp->nfsrecly_devid,
5678 				    NFSX_V4DEVICEID);
5679 		}
5680 	}
5681 	if (rp == NULL) {
5682 		if (orp == NULL)
5683 			LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5684 			    nfsrecly_list);
5685 		else
5686 			LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5687 	}
5688 	lyp->nfsly_flags |= NFSLY_RECALL;
5689 	wakeup(lyp->nfsly_clp);
5690 	return (0);
5691 }
5692 
5693 /*
5694  * Compare the two seqids for ordering. The trick is that the seqids can
5695  * wrap around from 0xffffffff->0, so check for the cases where one
5696  * has wrapped around.
5697  * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5698  */
5699 static int
5700 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5701 {
5702 
5703 	if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5704 		/* seqid2 has wrapped around. */
5705 		return (0);
5706 	if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5707 		/* seqid1 has wrapped around. */
5708 		return (1);
5709 	if (seqid1 <= seqid2)
5710 		return (1);
5711 	return (0);
5712 }
5713 
5714 /*
5715  * Do a layout return for each of the recalls.
5716  */
5717 static void
5718 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5719     struct ucred *cred, NFSPROC_T *p)
5720 {
5721 	struct nfsclrecalllayout *rp;
5722 	nfsv4stateid_t stateid;
5723 	int layouttype;
5724 
5725 	NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5726 	stateid.seqid = lyp->nfsly_stateid.seqid;
5727 	if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5728 		layouttype = NFSLAYOUT_NFSV4_1_FILES;
5729 	else
5730 		layouttype = NFSLAYOUT_FLEXFILE;
5731 	LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5732 		(void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5733 		    lyp->nfsly_fhlen, 0, layouttype,
5734 		    rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5735 		    rp->nfsrecly_off, rp->nfsrecly_len,
5736 		    &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5737 		    rp->nfsrecly_devid);
5738 	}
5739 }
5740 
5741 /*
5742  * Do the layout commit for a file layout.
5743  */
5744 static void
5745 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5746     struct ucred *cred, NFSPROC_T *p)
5747 {
5748 	struct nfsclflayout *flp;
5749 	uint64_t len;
5750 	int error, layouttype;
5751 
5752 	if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5753 		layouttype = NFSLAYOUT_NFSV4_1_FILES;
5754 	else
5755 		layouttype = NFSLAYOUT_FLEXFILE;
5756 	LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5757 		if (layouttype == NFSLAYOUT_FLEXFILE &&
5758 		    (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5759 			NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5760 			/* If not supported, don't bother doing it. */
5761 			NFSLOCKMNT(nmp);
5762 			nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5763 			NFSUNLOCKMNT(nmp);
5764 			break;
5765 		} else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5766 			len = flp->nfsfl_end - flp->nfsfl_off;
5767 			error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5768 			    lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5769 			    lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5770 			    layouttype, cred, p, NULL);
5771 			NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5772 			if (error == NFSERR_NOTSUPP) {
5773 				/* If not supported, don't bother doing it. */
5774 				NFSLOCKMNT(nmp);
5775 				nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5776 				NFSUNLOCKMNT(nmp);
5777 				break;
5778 			}
5779 		}
5780 	}
5781 }
5782 
5783 /*
5784  * Commit all layouts for a file (vnode).
5785  */
5786 int
5787 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5788 {
5789 	struct nfsclclient *clp;
5790 	struct nfscllayout *lyp;
5791 	struct nfsnode *np = VTONFS(vp);
5792 	mount_t mp;
5793 	struct nfsmount *nmp;
5794 
5795 	mp = vp->v_mount;
5796 	nmp = VFSTONFS(mp);
5797 	if (NFSHASNOLAYOUTCOMMIT(nmp))
5798 		return (0);
5799 	NFSLOCKCLSTATE();
5800 	clp = nmp->nm_clp;
5801 	if (clp == NULL) {
5802 		NFSUNLOCKCLSTATE();
5803 		return (EPERM);
5804 	}
5805 	lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5806 	if (lyp == NULL) {
5807 		NFSUNLOCKCLSTATE();
5808 		return (EPERM);
5809 	}
5810 	nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5811 	if (NFSCL_FORCEDISM(mp)) {
5812 		NFSUNLOCKCLSTATE();
5813 		return (EPERM);
5814 	}
5815 tryagain:
5816 	if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5817 		lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5818 		NFSUNLOCKCLSTATE();
5819 		NFSCL_DEBUG(4, "do layoutcommit2\n");
5820 		nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5821 		NFSLOCKCLSTATE();
5822 		goto tryagain;
5823 	}
5824 	nfsv4_relref(&lyp->nfsly_lock);
5825 	NFSUNLOCKCLSTATE();
5826 	return (0);
5827 }
5828