xref: /freebsd/sys/fs/nfsclient/nfs_clstate.c (revision a64729f5077d77e13b9497cb33ecb3c82e606ee8)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2009 Rick Macklem, University of Guelph
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  */
29 
30 #include <sys/cdefs.h>
31 /*
32  * These functions implement the client side state handling for NFSv4.
33  * NFSv4 state handling:
34  * - A lockowner is used to determine lock contention, so it
35  *   corresponds directly to a Posix pid. (1 to 1 mapping)
36  * - The correct granularity of an OpenOwner is not nearly so
37  *   obvious. An OpenOwner does the following:
38  *   - provides a serial sequencing of Open/Close/Lock-with-new-lockowner
39  *   - is used to check for Open/Share contention (not applicable to
40  *     this client, since all Opens are Deny_None)
41  *   As such, I considered both extreme.
42  *   1 OpenOwner per ClientID - Simple to manage, but fully serializes
43  *   all Open, Close and Lock (with a new lockowner) Ops.
44  *   1 OpenOwner for each Open - This one results in an OpenConfirm for
45  *   every Open, for most servers.
46  *   So, I chose to use the same mapping as I did for LockOwnwers.
47  *   The main concern here is that you can end up with multiple Opens
48  *   for the same File Handle, but on different OpenOwners (opens
49  *   inherited from parents, grandparents...) and you do not know
50  *   which of these the vnodeop close applies to. This is handled by
51  *   delaying the Close Op(s) until all of the Opens have been closed.
52  *   (It is not yet obvious if this is the correct granularity.)
53  * - How the code handles serialization:
54  *   - For the ClientId, it uses an exclusive lock while getting its
55  *     SetClientId and during recovery. Otherwise, it uses a shared
56  *     lock via a reference count.
57  *   - For the rest of the data structures, it uses an SMP mutex
58  *     (once the nfs client is SMP safe) and doesn't sleep while
59  *     manipulating the linked lists.
60  *   - The serialization of Open/Close/Lock/LockU falls out in the
61  *     "wash", since OpenOwners and LockOwners are both mapped from
62  *     Posix pid. In other words, there is only one Posix pid using
63  *     any given owner, so that owner is serialized. (If you change
64  *     the granularity of the OpenOwner, then code must be added to
65  *     serialize Ops on the OpenOwner.)
66  * - When to get rid of OpenOwners and LockOwners.
67  *   - The function nfscl_cleanup_common() is executed after a process exits.
68  *     It goes through the client list looking for all Open and Lock Owners.
69  *     When one is found, it is marked "defunct" or in the case of
70  *     an OpenOwner without any Opens, freed.
71  *     The renew thread scans for defunct Owners and gets rid of them,
72  *     if it can. The LockOwners will also be deleted when the
73  *     associated Open is closed.
74  *   - If the LockU or Close Op(s) fail during close in a way
75  *     that could be recovered upon retry, they are relinked to the
76  *     ClientId's defunct open list and retried by the renew thread
77  *     until they succeed or an unmount/recovery occurs.
78  *     (Since we are done with them, they do not need to be recovered.)
79  */
80 
81 #include <fs/nfs/nfsport.h>
82 
83 /*
84  * Global variables
85  */
86 extern struct nfsstatsv1 nfsstatsv1;
87 extern struct nfsreqhead nfsd_reqq;
88 extern u_int32_t newnfs_false, newnfs_true;
89 extern int nfscl_debuglevel;
90 extern int nfscl_enablecallb;
91 extern int nfs_numnfscbd;
92 NFSREQSPINLOCK;
93 NFSCLSTATEMUTEX;
94 int nfscl_inited = 0;
95 struct nfsclhead nfsclhead;	/* Head of clientid list */
96 
97 static int nfscl_deleghighwater = NFSCLDELEGHIGHWATER;
98 static int nfscl_delegcnt = 0;
99 static int nfscl_layoutcnt = 0;
100 static int nfscl_getopen(struct nfsclownerhead *, struct nfsclopenhash *,
101     u_int8_t *, int, u_int8_t *, u_int8_t *, u_int32_t,
102     struct nfscllockowner **, struct nfsclopen **);
103 static bool nfscl_checkown(struct nfsclowner *, struct nfsclopen *, uint8_t *,
104     uint8_t *, struct nfscllockowner **, struct nfsclopen **,
105     struct nfsclopen **);
106 static void nfscl_clrelease(struct nfsclclient *);
107 static void nfscl_unlinkopen(struct nfsclopen *);
108 static void nfscl_cleanclient(struct nfsclclient *);
109 static void nfscl_expireclient(struct nfsclclient *, struct nfsmount *,
110     struct ucred *, NFSPROC_T *);
111 static int nfscl_expireopen(struct nfsclclient *, struct nfsclopen *,
112     struct nfsmount *, struct ucred *, NFSPROC_T *);
113 static void nfscl_recover(struct nfsclclient *, bool *, struct ucred *,
114     NFSPROC_T *);
115 static void nfscl_insertlock(struct nfscllockowner *, struct nfscllock *,
116     struct nfscllock *, int);
117 static int nfscl_updatelock(struct nfscllockowner *, struct nfscllock **,
118     struct nfscllock **, int);
119 static void nfscl_delegreturnall(struct nfsclclient *, NFSPROC_T *,
120     struct nfscldeleghead *);
121 static u_int32_t nfscl_nextcbident(void);
122 static mount_t nfscl_getmnt(int, uint8_t *, u_int32_t, struct nfsclclient **);
123 static struct nfsclclient *nfscl_getclnt(u_int32_t);
124 static struct nfsclclient *nfscl_getclntsess(uint8_t *);
125 static struct nfscldeleg *nfscl_finddeleg(struct nfsclclient *, u_int8_t *,
126     int);
127 static void nfscl_retoncloselayout(vnode_t, struct nfsclclient *, uint8_t *,
128     int, struct nfsclrecalllayout **, struct nfscllayout **);
129 static void nfscl_reldevinfo_locked(struct nfscldevinfo *);
130 static struct nfscllayout *nfscl_findlayout(struct nfsclclient *, u_int8_t *,
131     int);
132 static struct nfscldevinfo *nfscl_finddevinfo(struct nfsclclient *, uint8_t *);
133 static int nfscl_checkconflict(struct nfscllockownerhead *, struct nfscllock *,
134     u_int8_t *, struct nfscllock **);
135 static void nfscl_freealllocks(struct nfscllockownerhead *, int);
136 static int nfscl_localconflict(struct nfsclclient *, u_int8_t *, int,
137     struct nfscllock *, u_int8_t *, struct nfscldeleg *, struct nfscllock **);
138 static void nfscl_newopen(struct nfsclclient *, struct nfscldeleg *,
139     struct nfsclowner **, struct nfsclowner **, struct nfsclopen **,
140     struct nfsclopen **, u_int8_t *, u_int8_t *, int, struct ucred *, int *);
141 static int nfscl_moveopen(vnode_t , struct nfsclclient *,
142     struct nfsmount *, struct nfsclopen *, struct nfsclowner *,
143     struct nfscldeleg *, struct ucred *, NFSPROC_T *);
144 static void nfscl_totalrecall(struct nfsclclient *);
145 static int nfscl_relock(vnode_t , struct nfsclclient *, struct nfsmount *,
146     struct nfscllockowner *, struct nfscllock *, struct ucred *, NFSPROC_T *);
147 static int nfscl_tryopen(struct nfsmount *, vnode_t , u_int8_t *, int,
148     u_int8_t *, int, u_int32_t, struct nfsclopen *, u_int8_t *, int,
149     struct nfscldeleg **, int, u_int32_t, struct ucred *, NFSPROC_T *);
150 static int nfscl_trylock(struct nfsmount *, vnode_t , u_int8_t *,
151     int, struct nfscllockowner *, int, int, u_int64_t, u_int64_t, short,
152     struct ucred *, NFSPROC_T *);
153 static int nfsrpc_reopen(struct nfsmount *, u_int8_t *, int, u_int32_t,
154     struct nfsclopen *, struct nfscldeleg **, struct ucred *, NFSPROC_T *);
155 static void nfscl_freedeleg(struct nfscldeleghead *, struct nfscldeleg *,
156     bool);
157 static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
158 static void nfscl_cleanup_common(struct nfsclclient *, u_int8_t *);
159 static int nfscl_recalldeleg(struct nfsclclient *, struct nfsmount *,
160     struct nfscldeleg *, vnode_t, struct ucred *, NFSPROC_T *, int,
161     vnode_t *);
162 static void nfscl_freeopenowner(struct nfsclowner *, int);
163 static void nfscl_cleandeleg(struct nfscldeleg *);
164 static void nfscl_emptylockowner(struct nfscllockowner *,
165     struct nfscllockownerfhhead *);
166 static void nfscl_mergeflayouts(struct nfsclflayouthead *,
167     struct nfsclflayouthead *);
168 static int nfscl_layoutrecall(int, struct nfscllayout *, uint32_t, uint64_t,
169     uint64_t, uint32_t, uint32_t, uint32_t, char *, struct nfsclrecalllayout *);
170 static int nfscl_seq(uint32_t, uint32_t);
171 static void nfscl_layoutreturn(struct nfsmount *, struct nfscllayout *,
172     struct ucred *, NFSPROC_T *);
173 static void nfscl_dolayoutcommit(struct nfsmount *, struct nfscllayout *,
174     struct ucred *, NFSPROC_T *);
175 
176 static short nfscberr_null[] = {
177 	0,
178 	0,
179 };
180 
181 static short nfscberr_getattr[] = {
182 	NFSERR_RESOURCE,
183 	NFSERR_BADHANDLE,
184 	NFSERR_BADXDR,
185 	NFSERR_RESOURCE,
186 	NFSERR_SERVERFAULT,
187 	0,
188 };
189 
190 static short nfscberr_recall[] = {
191 	NFSERR_RESOURCE,
192 	NFSERR_BADHANDLE,
193 	NFSERR_BADSTATEID,
194 	NFSERR_BADXDR,
195 	NFSERR_RESOURCE,
196 	NFSERR_SERVERFAULT,
197 	0,
198 };
199 
200 static short *nfscl_cberrmap[] = {
201 	nfscberr_null,
202 	nfscberr_null,
203 	nfscberr_null,
204 	nfscberr_getattr,
205 	nfscberr_recall
206 };
207 
208 #define	NETFAMILY(clp) \
209 		(((clp)->nfsc_flags & NFSCLFLAGS_AFINET6) ? AF_INET6 : AF_INET)
210 
211 /*
212  * Called for an open operation.
213  * If the nfhp argument is NULL, just get an openowner.
214  */
215 int
216 nfscl_open(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t amode, int usedeleg,
217     struct ucred *cred, NFSPROC_T *p, struct nfsclowner **owpp,
218     struct nfsclopen **opp, int *newonep, int *retp, int lockit, bool firstref)
219 {
220 	struct nfsclclient *clp;
221 	struct nfsclowner *owp, *nowp;
222 	struct nfsclopen *op = NULL, *nop = NULL;
223 	struct nfscldeleg *dp;
224 	struct nfsclownerhead *ohp;
225 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
226 	int ret;
227 
228 	if (newonep != NULL)
229 		*newonep = 0;
230 	if (opp != NULL)
231 		*opp = NULL;
232 	if (owpp != NULL)
233 		*owpp = NULL;
234 
235 	/*
236 	 * Might need one or both of these, so MALLOC them now, to
237 	 * avoid a tsleep() in MALLOC later.
238 	 */
239 	nowp = malloc(sizeof (struct nfsclowner),
240 	    M_NFSCLOWNER, M_WAITOK);
241 	if (nfhp != NULL) {
242 	    nop = malloc(sizeof (struct nfsclopen) +
243 		fhlen - 1, M_NFSCLOPEN, M_WAITOK);
244 	    nop->nfso_hash.le_prev = NULL;
245 	}
246 	ret = nfscl_getcl(vp->v_mount, cred, p, false, firstref, &clp);
247 	if (ret != 0) {
248 		free(nowp, M_NFSCLOWNER);
249 		if (nop != NULL)
250 			free(nop, M_NFSCLOPEN);
251 		return (ret);
252 	}
253 
254 	/*
255 	 * Get the Open iff it already exists.
256 	 * If none found, add the new one or return error, depending upon
257 	 * "create".
258 	 */
259 	NFSLOCKCLSTATE();
260 	dp = NULL;
261 	/* First check the delegation list */
262 	if (nfhp != NULL && usedeleg) {
263 		LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
264 			if (dp->nfsdl_fhlen == fhlen &&
265 			    !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
266 				if (!(amode & NFSV4OPEN_ACCESSWRITE) ||
267 				    (dp->nfsdl_flags & NFSCLDL_WRITE))
268 					break;
269 				dp = NULL;
270 				break;
271 			}
272 		}
273 	}
274 
275 	/* For NFSv4.1/4.2 and this option, use a single open_owner. */
276 	if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
277 		nfscl_filllockowner(NULL, own, F_POSIX);
278 	else
279 		nfscl_filllockowner(p->td_proc, own, F_POSIX);
280 	if (dp != NULL)
281 		ohp = &dp->nfsdl_owner;
282 	else
283 		ohp = &clp->nfsc_owner;
284 	/* Now, search for an openowner */
285 	LIST_FOREACH(owp, ohp, nfsow_list) {
286 		if (!NFSBCMP(owp->nfsow_owner, own, NFSV4CL_LOCKNAMELEN))
287 			break;
288 	}
289 
290 	/*
291 	 * Create a new open, as required.
292 	 */
293 	nfscl_newopen(clp, dp, &owp, &nowp, &op, &nop, own, nfhp, fhlen,
294 	    cred, newonep);
295 
296 	/*
297 	 * Now, check the mode on the open and return the appropriate
298 	 * value.
299 	 */
300 	if (retp != NULL) {
301 		if (nfhp != NULL && dp != NULL && nop == NULL)
302 			/* new local open on delegation */
303 			*retp = NFSCLOPEN_SETCRED;
304 		else
305 			*retp = NFSCLOPEN_OK;
306 	}
307 	if (op != NULL && (amode & ~(op->nfso_mode))) {
308 		op->nfso_mode |= amode;
309 		if (retp != NULL && dp == NULL)
310 			*retp = NFSCLOPEN_DOOPEN;
311 	}
312 
313 	/*
314 	 * Serialize modifications to the open owner for multiple threads
315 	 * within the same process using a read/write sleep lock.
316 	 * For NFSv4.1 and a single OpenOwner, allow concurrent open operations
317 	 * by acquiring a shared lock.  The close operations still use an
318 	 * exclusive lock for this case.
319 	 */
320 	if (lockit != 0) {
321 		if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount))) {
322 			/*
323 			 * Get a shared lock on the OpenOwner, but first
324 			 * wait for any pending exclusive lock, so that the
325 			 * exclusive locker gets priority.
326 			 */
327 			nfsv4_lock(&owp->nfsow_rwlock, 0, NULL,
328 			    NFSCLSTATEMUTEXPTR, NULL);
329 			nfsv4_getref(&owp->nfsow_rwlock, NULL,
330 			    NFSCLSTATEMUTEXPTR, NULL);
331 		} else
332 			nfscl_lockexcl(&owp->nfsow_rwlock, NFSCLSTATEMUTEXPTR);
333 	}
334 	NFSUNLOCKCLSTATE();
335 	if (nowp != NULL)
336 		free(nowp, M_NFSCLOWNER);
337 	if (nop != NULL)
338 		free(nop, M_NFSCLOPEN);
339 	if (owpp != NULL)
340 		*owpp = owp;
341 	if (opp != NULL)
342 		*opp = op;
343 	return (0);
344 }
345 
346 /*
347  * Create a new open, as required.
348  */
349 static void
350 nfscl_newopen(struct nfsclclient *clp, struct nfscldeleg *dp,
351     struct nfsclowner **owpp, struct nfsclowner **nowpp, struct nfsclopen **opp,
352     struct nfsclopen **nopp, u_int8_t *own, u_int8_t *fhp, int fhlen,
353     struct ucred *cred, int *newonep)
354 {
355 	struct nfsclowner *owp = *owpp, *nowp;
356 	struct nfsclopen *op, *nop;
357 
358 	if (nowpp != NULL)
359 		nowp = *nowpp;
360 	else
361 		nowp = NULL;
362 	if (nopp != NULL)
363 		nop = *nopp;
364 	else
365 		nop = NULL;
366 	if (owp == NULL && nowp != NULL) {
367 		NFSBCOPY(own, nowp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
368 		LIST_INIT(&nowp->nfsow_open);
369 		nowp->nfsow_clp = clp;
370 		nowp->nfsow_seqid = 0;
371 		nowp->nfsow_defunct = 0;
372 		nfscl_lockinit(&nowp->nfsow_rwlock);
373 		if (dp != NULL) {
374 			nfsstatsv1.cllocalopenowners++;
375 			LIST_INSERT_HEAD(&dp->nfsdl_owner, nowp, nfsow_list);
376 		} else {
377 			nfsstatsv1.clopenowners++;
378 			LIST_INSERT_HEAD(&clp->nfsc_owner, nowp, nfsow_list);
379 		}
380 		owp = *owpp = nowp;
381 		*nowpp = NULL;
382 		if (newonep != NULL)
383 			*newonep = 1;
384 	}
385 
386 	 /* If an fhp has been specified, create an Open as well. */
387 	if (fhp != NULL) {
388 		/* and look for the correct open, based upon FH */
389 		LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
390 			if (op->nfso_fhlen == fhlen &&
391 			    !NFSBCMP(op->nfso_fh, fhp, fhlen))
392 				break;
393 		}
394 		if (op == NULL && nop != NULL) {
395 			nop->nfso_own = owp;
396 			nop->nfso_mode = 0;
397 			nop->nfso_opencnt = 0;
398 			nop->nfso_posixlock = 1;
399 			nop->nfso_fhlen = fhlen;
400 			NFSBCOPY(fhp, nop->nfso_fh, fhlen);
401 			LIST_INIT(&nop->nfso_lock);
402 			nop->nfso_stateid.seqid = 0;
403 			nop->nfso_stateid.other[0] = 0;
404 			nop->nfso_stateid.other[1] = 0;
405 			nop->nfso_stateid.other[2] = 0;
406 			KASSERT(cred != NULL, ("%s: cred NULL\n", __func__));
407 			newnfs_copyincred(cred, &nop->nfso_cred);
408 			if (dp != NULL) {
409 				TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
410 				TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
411 				    nfsdl_list);
412 				dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
413 				nfsstatsv1.cllocalopens++;
414 			} else {
415 				LIST_INSERT_HEAD(NFSCLOPENHASH(clp, fhp, fhlen),
416 				    nop, nfso_hash);
417 				nfsstatsv1.clopens++;
418 			}
419 			LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
420 			*opp = nop;
421 			*nopp = NULL;
422 			if (newonep != NULL)
423 				*newonep = 1;
424 		} else {
425 			*opp = op;
426 		}
427 	}
428 }
429 
430 /*
431  * Called to find/add a delegation to a client.
432  */
433 int
434 nfscl_deleg(mount_t mp, struct nfsclclient *clp, u_int8_t *nfhp,
435     int fhlen, struct ucred *cred, NFSPROC_T *p, struct nfscldeleg **dpp)
436 {
437 	struct nfscldeleg *dp = *dpp, *tdp;
438 	struct nfsmount *nmp;
439 
440 	KASSERT(mp != NULL, ("nfscl_deleg: mp NULL"));
441 	nmp = VFSTONFS(mp);
442 
443 	/*
444 	 * Since a delegation might be added to the mount,
445 	 * set NFSMNTP_DELEGISSUED now.  If a delegation already
446 	 * exagain ists, setting this flag is harmless.
447 	 */
448 	NFSLOCKMNT(nmp);
449 	nmp->nm_privflag |= NFSMNTP_DELEGISSUED;
450 	NFSUNLOCKMNT(nmp);
451 
452 	/* Look for the correct deleg, based upon FH */
453 	NFSLOCKCLSTATE();
454 	tdp = nfscl_finddeleg(clp, nfhp, fhlen);
455 	if (tdp == NULL) {
456 		if (dp == NULL) {
457 			NFSUNLOCKCLSTATE();
458 			return (NFSERR_BADSTATEID);
459 		}
460 		*dpp = NULL;
461 		TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
462 		LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
463 		    nfsdl_hash);
464 		dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
465 		nfsstatsv1.cldelegates++;
466 		nfscl_delegcnt++;
467 	} else {
468 		/*
469 		 * A delegation already exists.  If the new one is a Write
470 		 * delegation and the old one a Read delegation, return the
471 		 * Read delegation.  Otherwise, return the new delegation.
472 		 */
473 		if (dp != NULL) {
474 			if ((dp->nfsdl_flags & NFSCLDL_WRITE) != 0 &&
475 			    (tdp->nfsdl_flags & NFSCLDL_READ) != 0) {
476 				TAILQ_REMOVE(&clp->nfsc_deleg, tdp, nfsdl_list);
477 				LIST_REMOVE(tdp, nfsdl_hash);
478 				*dpp = NULL;
479 				TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
480 				    nfsdl_list);
481 				LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp,
482 				    fhlen), dp, nfsdl_hash);
483 				dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
484 			} else {
485 				*dpp = NULL;
486 				tdp = dp;	/* Return this one. */
487 			}
488 		} else {
489 			*dpp = tdp;
490 			tdp = NULL;
491 		}
492 	}
493 	NFSUNLOCKCLSTATE();
494 	if (tdp != NULL) {
495 		nfscl_trydelegreturn(tdp, cred, nmp, p);
496 		free(tdp, M_NFSCLDELEG);
497 	}
498 	return (0);
499 }
500 
501 /*
502  * Find a delegation for this file handle. Return NULL upon failure.
503  */
504 static struct nfscldeleg *
505 nfscl_finddeleg(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
506 {
507 	struct nfscldeleg *dp;
508 
509 	LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
510 	    if (dp->nfsdl_fhlen == fhlen &&
511 		!NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
512 		break;
513 	}
514 	return (dp);
515 }
516 
517 /*
518  * Get a stateid for an I/O operation. First, look for an open and iff
519  * found, return either a lockowner stateid or the open stateid.
520  * If no Open is found, just return error and the special stateid of all zeros.
521  */
522 int
523 nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
524     int fords, struct ucred *cred, NFSPROC_T *p, nfsv4stateid_t *stateidp,
525     void **lckpp)
526 {
527 	struct nfsclclient *clp;
528 	struct nfsclopen *op = NULL, *top;
529 	struct nfsclopenhash *oph;
530 	struct nfscllockowner *lp;
531 	struct nfscldeleg *dp;
532 	struct nfsnode *np;
533 	struct nfsmount *nmp;
534 	struct nfscred ncr;
535 	u_int8_t own[NFSV4CL_LOCKNAMELEN], lockown[NFSV4CL_LOCKNAMELEN];
536 	int error;
537 	bool done;
538 
539 	*lckpp = NULL;
540 	/*
541 	 * Initially, just set the special stateid of all zeros.
542 	 * (Don't do this for a DS, since the special stateid can't be used.)
543 	 */
544 	if (fords == 0) {
545 		stateidp->seqid = 0;
546 		stateidp->other[0] = 0;
547 		stateidp->other[1] = 0;
548 		stateidp->other[2] = 0;
549 	}
550 	if (vp->v_type != VREG)
551 		return (EISDIR);
552 	np = VTONFS(vp);
553 	nmp = VFSTONFS(vp->v_mount);
554 
555 	/*
556 	 * For "oneopenown" mounts, first check for a cached open in the
557 	 * NFS vnode, that can be used as a stateid.  This can only be
558 	 * done if no delegations have been issued to the mount and no
559 	 * byte range file locking has been done for the file.
560 	 */
561 	if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && fords == 0) {
562 		NFSLOCKMNT(nmp);
563 		NFSLOCKNODE(np);
564 		if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 &&
565 		    (np->n_flag & NMIGHTBELOCKED) == 0 &&
566 		    np->n_openstateid != NULL) {
567 			stateidp->seqid = 0;
568 			stateidp->other[0] =
569 			    np->n_openstateid->nfso_stateid.other[0];
570 			stateidp->other[1] =
571 			    np->n_openstateid->nfso_stateid.other[1];
572 			stateidp->other[2] =
573 			    np->n_openstateid->nfso_stateid.other[2];
574 			NFSUNLOCKNODE(np);
575 			NFSUNLOCKMNT(nmp);
576 			return (0);
577 		}
578 		NFSUNLOCKNODE(np);
579 		NFSUNLOCKMNT(nmp);
580 	}
581 
582 	NFSLOCKCLSTATE();
583 	clp = nfscl_findcl(nmp);
584 	if (clp == NULL) {
585 		NFSUNLOCKCLSTATE();
586 		return (EACCES);
587 	}
588 
589 	/*
590 	 * Wait for recovery to complete.
591 	 */
592 	while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
593 		(void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
594 		    PZERO, "nfsrecvr", NULL);
595 
596 	/*
597 	 * First, look for a delegation.
598 	 */
599 	LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
600 		if (dp->nfsdl_fhlen == fhlen &&
601 		    !NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
602 			if (!(mode & NFSV4OPEN_ACCESSWRITE) ||
603 			    (dp->nfsdl_flags & NFSCLDL_WRITE)) {
604 				if (NFSHASNFSV4N(nmp))
605 					stateidp->seqid = 0;
606 				else
607 					stateidp->seqid =
608 					    dp->nfsdl_stateid.seqid;
609 				stateidp->other[0] = dp->nfsdl_stateid.other[0];
610 				stateidp->other[1] = dp->nfsdl_stateid.other[1];
611 				stateidp->other[2] = dp->nfsdl_stateid.other[2];
612 				if (!(np->n_flag & NDELEGRECALL)) {
613 					TAILQ_REMOVE(&clp->nfsc_deleg, dp,
614 					    nfsdl_list);
615 					TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
616 					    nfsdl_list);
617 					dp->nfsdl_timestamp = NFSD_MONOSEC +
618 					    120;
619 					dp->nfsdl_rwlock.nfslock_usecnt++;
620 					*lckpp = (void *)&dp->nfsdl_rwlock;
621 				}
622 				NFSUNLOCKCLSTATE();
623 				return (0);
624 			}
625 			break;
626 		}
627 	}
628 
629 	if (p != NULL) {
630 		/*
631 		 * If p != NULL, we want to search the parentage tree
632 		 * for a matching OpenOwner and use that.
633 		 */
634 		if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
635 			nfscl_filllockowner(NULL, own, F_POSIX);
636 		else
637 			nfscl_filllockowner(p->td_proc, own, F_POSIX);
638 		nfscl_filllockowner(p->td_proc, lockown, F_POSIX);
639 		lp = NULL;
640 		error = nfscl_getopen(NULL, clp->nfsc_openhash, nfhp, fhlen,
641 		    own, lockown, mode, &lp, &op);
642 		if (error == 0 && lp != NULL && fords == 0) {
643 			/* Don't return a lock stateid for a DS. */
644 			if (NFSHASNFSV4N(nmp))
645 				stateidp->seqid = 0;
646 			else
647 				stateidp->seqid = lp->nfsl_stateid.seqid;
648 			stateidp->other[0] =
649 			    lp->nfsl_stateid.other[0];
650 			stateidp->other[1] =
651 			    lp->nfsl_stateid.other[1];
652 			stateidp->other[2] =
653 			    lp->nfsl_stateid.other[2];
654 			NFSUNLOCKCLSTATE();
655 			return (0);
656 		}
657 	}
658 	if (op == NULL) {
659 		/* If not found, just look for any OpenOwner that will work. */
660 		top = NULL;
661 		done = false;
662 		oph = NFSCLOPENHASH(clp, nfhp, fhlen);
663 		LIST_FOREACH(op, oph, nfso_hash) {
664 			if (op->nfso_fhlen == fhlen &&
665 			    !NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
666 				if (top == NULL && (op->nfso_mode &
667 				    NFSV4OPEN_ACCESSWRITE) != 0 &&
668 				    (mode & NFSV4OPEN_ACCESSREAD) != 0)
669 					top = op;
670 				if ((mode & op->nfso_mode) == mode) {
671 					/* LRU order the hash list. */
672 					LIST_REMOVE(op, nfso_hash);
673 					LIST_INSERT_HEAD(oph, op, nfso_hash);
674 					done = true;
675 					break;
676 				}
677 			}
678 		}
679 		if (!done) {
680 			NFSCL_DEBUG(2, "openmode top=%p\n", top);
681 			if (top == NULL || NFSHASOPENMODE(nmp)) {
682 				NFSUNLOCKCLSTATE();
683 				return (ENOENT);
684 			} else
685 				op = top;
686 		}
687 		/*
688 		 * For read aheads or write behinds, use the open cred.
689 		 * A read ahead or write behind is indicated by p == NULL.
690 		 */
691 		if (p == NULL)
692 			memcpy(&ncr, &op->nfso_cred, sizeof(ncr));
693 	}
694 
695 	/*
696 	 * No lock stateid, so return the open stateid.
697 	 */
698 	if (NFSHASNFSV4N(nmp))
699 		stateidp->seqid = 0;
700 	else
701 		stateidp->seqid = op->nfso_stateid.seqid;
702 	stateidp->other[0] = op->nfso_stateid.other[0];
703 	stateidp->other[1] = op->nfso_stateid.other[1];
704 	stateidp->other[2] = op->nfso_stateid.other[2];
705 	NFSUNLOCKCLSTATE();
706 	if (p == NULL)
707 		newnfs_copycred(&ncr, cred);
708 	return (0);
709 }
710 
711 /*
712  * Search for a matching file, mode and, optionally, lockowner.
713  */
714 static int
715 nfscl_getopen(struct nfsclownerhead *ohp, struct nfsclopenhash *ohashp,
716     u_int8_t *nfhp, int fhlen, u_int8_t *openown, u_int8_t *lockown,
717     u_int32_t mode, struct nfscllockowner **lpp, struct nfsclopen **opp)
718 {
719 	struct nfsclowner *owp;
720 	struct nfsclopen *op, *rop, *rop2;
721 	struct nfsclopenhash *oph;
722 	bool keep_looping;
723 
724 	KASSERT(ohp == NULL || ohashp == NULL, ("nfscl_getopen: "
725 	    "only one of ohp and ohashp can be set"));
726 	if (lpp != NULL)
727 		*lpp = NULL;
728 	/*
729 	 * rop will be set to the open to be returned. There are three
730 	 * variants of this, all for an open of the correct file:
731 	 * 1 - A match of lockown.
732 	 * 2 - A match of the openown, when no lockown match exists.
733 	 * 3 - A match for any open, if no openown or lockown match exists.
734 	 * Looking for #2 over #3 probably isn't necessary, but since
735 	 * RFC3530 is vague w.r.t. the relationship between openowners and
736 	 * lockowners, I think this is the safer way to go.
737 	 */
738 	rop = NULL;
739 	rop2 = NULL;
740 	keep_looping = true;
741 	/* Search the client list */
742 	if (ohashp == NULL) {
743 		/* Search the local opens on the delegation. */
744 		LIST_FOREACH(owp, ohp, nfsow_list) {
745 			/* and look for the correct open */
746 			LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
747 				if (op->nfso_fhlen == fhlen &&
748 				    !NFSBCMP(op->nfso_fh, nfhp, fhlen)
749 				    && (op->nfso_mode & mode) == mode)
750 					keep_looping = nfscl_checkown(owp, op, openown,
751 					    lockown, lpp, &rop, &rop2);
752 				if (!keep_looping)
753 					break;
754 			}
755 			if (!keep_looping)
756 				break;
757 		}
758 	} else {
759 		/* Search for matching opens on the hash list. */
760 		oph = &ohashp[NFSCLOPENHASHFUNC(nfhp, fhlen)];
761 		LIST_FOREACH(op, oph, nfso_hash) {
762 			if (op->nfso_fhlen == fhlen &&
763 			    !NFSBCMP(op->nfso_fh, nfhp, fhlen)
764 			    && (op->nfso_mode & mode) == mode)
765 				keep_looping = nfscl_checkown(op->nfso_own, op,
766 				    openown, lockown, lpp, &rop, &rop2);
767 			if (!keep_looping) {
768 				/* LRU order the hash list. */
769 				LIST_REMOVE(op, nfso_hash);
770 				LIST_INSERT_HEAD(oph, op, nfso_hash);
771 				break;
772 			}
773 		}
774 	}
775 	if (rop == NULL)
776 		rop = rop2;
777 	if (rop == NULL)
778 		return (EBADF);
779 	*opp = rop;
780 	return (0);
781 }
782 
783 /* Check for an owner match. */
784 static bool
785 nfscl_checkown(struct nfsclowner *owp, struct nfsclopen *op, uint8_t *openown,
786     uint8_t *lockown, struct nfscllockowner **lpp, struct nfsclopen **ropp,
787     struct nfsclopen **ropp2)
788 {
789 	struct nfscllockowner *lp;
790 	bool keep_looping;
791 
792 	keep_looping = true;
793 	if (lpp != NULL) {
794 		/* Now look for a matching lockowner. */
795 		LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
796 			if (!NFSBCMP(lp->nfsl_owner, lockown,
797 			    NFSV4CL_LOCKNAMELEN)) {
798 				*lpp = lp;
799 				*ropp = op;
800 				return (false);
801 			}
802 		}
803 	}
804 	if (*ropp == NULL && !NFSBCMP(owp->nfsow_owner, openown,
805 	    NFSV4CL_LOCKNAMELEN)) {
806 		*ropp = op;
807 		if (lpp == NULL)
808 			keep_looping = false;
809 	}
810 	if (*ropp2 == NULL)
811 		*ropp2 = op;
812 	return (keep_looping);
813 }
814 
815 /*
816  * Release use of an open owner. Called when open operations are done
817  * with the open owner.
818  */
819 void
820 nfscl_ownerrelease(struct nfsmount *nmp, struct nfsclowner *owp,
821     __unused int error, __unused int candelete, int unlocked)
822 {
823 
824 	if (owp == NULL)
825 		return;
826 	NFSLOCKCLSTATE();
827 	if (unlocked == 0) {
828 		if (NFSHASONEOPENOWN(nmp))
829 			nfsv4_relref(&owp->nfsow_rwlock);
830 		else
831 			nfscl_lockunlock(&owp->nfsow_rwlock);
832 	}
833 	nfscl_clrelease(owp->nfsow_clp);
834 	NFSUNLOCKCLSTATE();
835 }
836 
837 /*
838  * Release use of an open structure under an open owner.
839  */
840 void
841 nfscl_openrelease(struct nfsmount *nmp, struct nfsclopen *op, int error,
842     int candelete)
843 {
844 	struct nfsclclient *clp;
845 	struct nfsclowner *owp;
846 
847 	if (op == NULL)
848 		return;
849 	NFSLOCKCLSTATE();
850 	owp = op->nfso_own;
851 	if (NFSHASONEOPENOWN(nmp))
852 		nfsv4_relref(&owp->nfsow_rwlock);
853 	else
854 		nfscl_lockunlock(&owp->nfsow_rwlock);
855 	clp = owp->nfsow_clp;
856 	if (error && candelete && op->nfso_opencnt == 0)
857 		nfscl_freeopen(op, 0, true);
858 	nfscl_clrelease(clp);
859 	NFSUNLOCKCLSTATE();
860 }
861 
862 /*
863  * Called to get a clientid structure. It will optionally lock the
864  * client data structures to do the SetClientId/SetClientId_confirm,
865  * but will release that lock and return the clientid with a reference
866  * count on it.
867  * If the "cred" argument is NULL, a new clientid should not be created.
868  * If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
869  * be done.
870  * It always clpp with a reference count on it, unless returning an error.
871  */
872 int
873 nfscl_getcl(struct mount *mp, struct ucred *cred, NFSPROC_T *p,
874     bool tryminvers, bool firstref, struct nfsclclient **clpp)
875 {
876 	struct nfsclclient *clp;
877 	struct nfsclclient *newclp = NULL;
878 	struct nfsmount *nmp;
879 	char uuid[HOSTUUIDLEN];
880 	int igotlock = 0, error, trystalecnt, clidinusedelay, i;
881 	u_int16_t idlen = 0;
882 
883 	nmp = VFSTONFS(mp);
884 	if (cred != NULL) {
885 		getcredhostuuid(cred, uuid, sizeof uuid);
886 		idlen = strlen(uuid);
887 		if (idlen > 0)
888 			idlen += sizeof (u_int64_t);
889 		else
890 			idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
891 		newclp = malloc(
892 		    sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
893 		    M_WAITOK | M_ZERO);
894 	}
895 	NFSLOCKCLSTATE();
896 	/*
897 	 * If a forced dismount is already in progress, don't
898 	 * allocate a new clientid and get out now. For the case where
899 	 * clp != NULL, this is a harmless optimization.
900 	 */
901 	if (NFSCL_FORCEDISM(mp)) {
902 		NFSUNLOCKCLSTATE();
903 		if (newclp != NULL)
904 			free(newclp, M_NFSCLCLIENT);
905 		return (EBADF);
906 	}
907 	clp = nmp->nm_clp;
908 	if (clp == NULL) {
909 		if (newclp == NULL) {
910 			NFSUNLOCKCLSTATE();
911 			return (EACCES);
912 		}
913 		clp = newclp;
914 		clp->nfsc_idlen = idlen;
915 		LIST_INIT(&clp->nfsc_owner);
916 		TAILQ_INIT(&clp->nfsc_deleg);
917 		TAILQ_INIT(&clp->nfsc_layout);
918 		LIST_INIT(&clp->nfsc_devinfo);
919 		for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
920 			LIST_INIT(&clp->nfsc_deleghash[i]);
921 		for (i = 0; i < NFSCLOPENHASHSIZE; i++)
922 			LIST_INIT(&clp->nfsc_openhash[i]);
923 		for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
924 			LIST_INIT(&clp->nfsc_layouthash[i]);
925 		clp->nfsc_flags = NFSCLFLAGS_INITED;
926 		clp->nfsc_clientidrev = 1;
927 		clp->nfsc_cbident = nfscl_nextcbident();
928 		nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
929 		    clp->nfsc_idlen);
930 		LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
931 		nmp->nm_clp = clp;
932 		clp->nfsc_nmp = nmp;
933 	} else {
934 		if (newclp != NULL)
935 			free(newclp, M_NFSCLCLIENT);
936 	}
937 	while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
938 	    !NFSCL_FORCEDISM(mp))
939 		igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
940 		    NFSCLSTATEMUTEXPTR, mp);
941 	if (igotlock == 0) {
942 		/*
943 		 * Call nfsv4_lock() with "iwantlock == 0" on the firstref so
944 		 * that it will wait for a pending exclusive lock request.
945 		 * This gives the exclusive lock request priority over this
946 		 * shared lock request.
947 		 * An exclusive lock on nfsc_lock is used mainly for server
948 		 * crash recoveries and delegation recalls.
949 		 */
950 		if (firstref)
951 			nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR,
952 			    mp);
953 		nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
954 	}
955 	if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
956 		/*
957 		 * Both nfsv4_lock() and nfsv4_getref() know to check
958 		 * for NFSCL_FORCEDISM() and return without sleeping to
959 		 * wait for the exclusive lock to be released, since it
960 		 * might be held by nfscl_umount() and we need to get out
961 		 * now for that case and not wait until nfscl_umount()
962 		 * releases it.
963 		 */
964 		NFSUNLOCKCLSTATE();
965 		return (EBADF);
966 	}
967 	NFSUNLOCKCLSTATE();
968 
969 	/*
970 	 * If it needs a clientid, do the setclientid now.
971 	 */
972 	if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
973 		if (!igotlock)
974 			panic("nfscl_clget");
975 		if (p == NULL || cred == NULL) {
976 			NFSLOCKCLSTATE();
977 			nfsv4_unlock(&clp->nfsc_lock, 0);
978 			NFSUNLOCKCLSTATE();
979 			return (EACCES);
980 		}
981 		/*
982 		 * If RFC3530 Sec. 14.2.33 is taken literally,
983 		 * NFSERR_CLIDINUSE will be returned persistently for the
984 		 * case where a new mount of the same file system is using
985 		 * a different principal. In practice, NFSERR_CLIDINUSE is
986 		 * only returned when there is outstanding unexpired state
987 		 * on the clientid. As such, try for twice the lease
988 		 * interval, if we know what that is. Otherwise, make a
989 		 * wild ass guess.
990 		 * The case of returning NFSERR_STALECLIENTID is far less
991 		 * likely, but might occur if there is a significant delay
992 		 * between doing the SetClientID and SetClientIDConfirm Ops,
993 		 * such that the server throws away the clientid before
994 		 * receiving the SetClientIDConfirm.
995 		 */
996 		if (clp->nfsc_renew > 0)
997 			clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
998 		else
999 			clidinusedelay = 120;
1000 		trystalecnt = 3;
1001 		do {
1002 			error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
1003 			if (error == NFSERR_STALECLIENTID ||
1004 			    error == NFSERR_STALEDONTRECOVER ||
1005 			    error == NFSERR_BADSESSION ||
1006 			    error == NFSERR_CLIDINUSE) {
1007 				(void) nfs_catnap(PZERO, error, "nfs_setcl");
1008 			} else if (error == NFSERR_MINORVERMISMATCH &&
1009 			    tryminvers) {
1010 				if (nmp->nm_minorvers > 0)
1011 					nmp->nm_minorvers--;
1012 				else
1013 					tryminvers = false;
1014 			}
1015 		} while (((error == NFSERR_STALECLIENTID ||
1016 		     error == NFSERR_BADSESSION ||
1017 		     error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) ||
1018 		    (error == NFSERR_CLIDINUSE && --clidinusedelay > 0) ||
1019 		    (error == NFSERR_MINORVERMISMATCH && tryminvers));
1020 		if (error) {
1021 			NFSLOCKCLSTATE();
1022 			nfsv4_unlock(&clp->nfsc_lock, 0);
1023 			NFSUNLOCKCLSTATE();
1024 			return (error);
1025 		}
1026 		clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
1027 	}
1028 	if (igotlock) {
1029 		NFSLOCKCLSTATE();
1030 		nfsv4_unlock(&clp->nfsc_lock, 1);
1031 		NFSUNLOCKCLSTATE();
1032 	}
1033 
1034 	*clpp = clp;
1035 	return (0);
1036 }
1037 
1038 /*
1039  * Get a reference to a clientid and return it, if valid.
1040  */
1041 struct nfsclclient *
1042 nfscl_findcl(struct nfsmount *nmp)
1043 {
1044 	struct nfsclclient *clp;
1045 
1046 	clp = nmp->nm_clp;
1047 	if (clp == NULL || !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
1048 		return (NULL);
1049 	return (clp);
1050 }
1051 
1052 /*
1053  * Release the clientid structure. It may be locked or reference counted.
1054  */
1055 static void
1056 nfscl_clrelease(struct nfsclclient *clp)
1057 {
1058 
1059 	if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
1060 		nfsv4_unlock(&clp->nfsc_lock, 0);
1061 	else
1062 		nfsv4_relref(&clp->nfsc_lock);
1063 }
1064 
1065 /*
1066  * External call for nfscl_clrelease.
1067  */
1068 void
1069 nfscl_clientrelease(struct nfsclclient *clp)
1070 {
1071 
1072 	NFSLOCKCLSTATE();
1073 	if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
1074 		nfsv4_unlock(&clp->nfsc_lock, 0);
1075 	else
1076 		nfsv4_relref(&clp->nfsc_lock);
1077 	NFSUNLOCKCLSTATE();
1078 }
1079 
1080 /*
1081  * Called when wanting to lock a byte region.
1082  */
1083 int
1084 nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1085     short type, struct ucred *cred, NFSPROC_T *p, struct nfsclclient *rclp,
1086     int recovery, void *id, int flags, u_int8_t *rownp, u_int8_t *ropenownp,
1087     struct nfscllockowner **lpp, int *newonep, int *donelocallyp)
1088 {
1089 	struct nfscllockowner *lp;
1090 	struct nfsclopen *op;
1091 	struct nfsclclient *clp;
1092 	struct nfscllockowner *nlp;
1093 	struct nfscllock *nlop, *otherlop;
1094 	struct nfscldeleg *dp = NULL, *ldp = NULL;
1095 	struct nfscllockownerhead *lhp = NULL;
1096 	struct nfsnode *np;
1097 	u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
1098 	u_int8_t *openownp;
1099 	int error = 0, ret, donelocally = 0;
1100 	u_int32_t mode;
1101 
1102 	/* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1103 	mode = 0;
1104 	np = VTONFS(vp);
1105 	*lpp = NULL;
1106 	lp = NULL;
1107 	*newonep = 0;
1108 	*donelocallyp = 0;
1109 
1110 	/*
1111 	 * Might need these, so MALLOC them now, to
1112 	 * avoid a tsleep() in MALLOC later.
1113 	 */
1114 	nlp = malloc(
1115 	    sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1116 	otherlop = malloc(
1117 	    sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1118 	nlop = malloc(
1119 	    sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1120 	nlop->nfslo_type = type;
1121 	nlop->nfslo_first = off;
1122 	if (len == NFS64BITSSET) {
1123 		nlop->nfslo_end = NFS64BITSSET;
1124 	} else {
1125 		nlop->nfslo_end = off + len;
1126 		if (nlop->nfslo_end <= nlop->nfslo_first)
1127 			error = NFSERR_INVAL;
1128 	}
1129 
1130 	if (!error) {
1131 		if (recovery)
1132 			clp = rclp;
1133 		else
1134 			error = nfscl_getcl(vp->v_mount, cred, p, false, true,
1135 			    &clp);
1136 	}
1137 	if (error) {
1138 		free(nlp, M_NFSCLLOCKOWNER);
1139 		free(otherlop, M_NFSCLLOCK);
1140 		free(nlop, M_NFSCLLOCK);
1141 		return (error);
1142 	}
1143 
1144 	op = NULL;
1145 	if (recovery) {
1146 		ownp = rownp;
1147 		openownp = ropenownp;
1148 	} else {
1149 		nfscl_filllockowner(id, own, flags);
1150 		ownp = own;
1151 		if (NFSHASONEOPENOWN(VFSTONFS(vp->v_mount)))
1152 			nfscl_filllockowner(NULL, openown, F_POSIX);
1153 		else
1154 			nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1155 		openownp = openown;
1156 	}
1157 	if (!recovery) {
1158 		NFSLOCKCLSTATE();
1159 		/*
1160 		 * First, search for a delegation. If one exists for this file,
1161 		 * the lock can be done locally against it, so long as there
1162 		 * isn't a local lock conflict.
1163 		 */
1164 		ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1165 		    np->n_fhp->nfh_len);
1166 		/* Just sanity check for correct type of delegation */
1167 		if (dp != NULL && ((dp->nfsdl_flags &
1168 		    (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) != 0 ||
1169 		     (type == F_WRLCK &&
1170 		      (dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1171 			dp = NULL;
1172 	}
1173 	if (dp != NULL) {
1174 		/* Now, find an open and maybe a lockowner. */
1175 		ret = nfscl_getopen(&dp->nfsdl_owner, NULL, np->n_fhp->nfh_fh,
1176 		    np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1177 		if (ret)
1178 			ret = nfscl_getopen(NULL, clp->nfsc_openhash,
1179 			    np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1180 			    ownp, mode, NULL, &op);
1181 		if (!ret) {
1182 			lhp = &dp->nfsdl_lock;
1183 			TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1184 			TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1185 			dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1186 			donelocally = 1;
1187 		} else {
1188 			dp = NULL;
1189 		}
1190 	}
1191 	if (!donelocally) {
1192 		/*
1193 		 * Get the related Open and maybe lockowner.
1194 		 */
1195 		error = nfscl_getopen(NULL, clp->nfsc_openhash,
1196 		    np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1197 		    ownp, mode, &lp, &op);
1198 		if (!error)
1199 			lhp = &op->nfso_lock;
1200 	}
1201 	if (!error && !recovery)
1202 		error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1203 		    np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1204 	if (error) {
1205 		if (!recovery) {
1206 			nfscl_clrelease(clp);
1207 			NFSUNLOCKCLSTATE();
1208 		}
1209 		free(nlp, M_NFSCLLOCKOWNER);
1210 		free(otherlop, M_NFSCLLOCK);
1211 		free(nlop, M_NFSCLLOCK);
1212 		return (error);
1213 	}
1214 
1215 	/*
1216 	 * Ok, see if a lockowner exists and create one, as required.
1217 	 */
1218 	if (lp == NULL)
1219 		LIST_FOREACH(lp, lhp, nfsl_list) {
1220 			if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1221 				break;
1222 		}
1223 	if (lp == NULL) {
1224 		NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1225 		if (recovery)
1226 			NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1227 			    NFSV4CL_LOCKNAMELEN);
1228 		else
1229 			NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1230 			    NFSV4CL_LOCKNAMELEN);
1231 		nlp->nfsl_seqid = 0;
1232 		nlp->nfsl_lockflags = flags;
1233 		nlp->nfsl_inprog = NULL;
1234 		nfscl_lockinit(&nlp->nfsl_rwlock);
1235 		LIST_INIT(&nlp->nfsl_lock);
1236 		if (donelocally) {
1237 			nlp->nfsl_open = NULL;
1238 			nfsstatsv1.cllocallockowners++;
1239 		} else {
1240 			nlp->nfsl_open = op;
1241 			nfsstatsv1.cllockowners++;
1242 		}
1243 		LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1244 		lp = nlp;
1245 		nlp = NULL;
1246 		*newonep = 1;
1247 	}
1248 
1249 	/*
1250 	 * Now, update the byte ranges for locks.
1251 	 */
1252 	ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1253 	if (!ret)
1254 		donelocally = 1;
1255 	if (donelocally) {
1256 		*donelocallyp = 1;
1257 		if (!recovery)
1258 			nfscl_clrelease(clp);
1259 	} else {
1260 		/*
1261 		 * Serial modifications on the lock owner for multiple threads
1262 		 * for the same process using a read/write lock.
1263 		 */
1264 		if (!recovery)
1265 			nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1266 	}
1267 	if (!recovery)
1268 		NFSUNLOCKCLSTATE();
1269 
1270 	if (nlp)
1271 		free(nlp, M_NFSCLLOCKOWNER);
1272 	if (nlop)
1273 		free(nlop, M_NFSCLLOCK);
1274 	if (otherlop)
1275 		free(otherlop, M_NFSCLLOCK);
1276 
1277 	*lpp = lp;
1278 	return (0);
1279 }
1280 
1281 /*
1282  * Called to unlock a byte range, for LockU.
1283  */
1284 int
1285 nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1286     __unused struct ucred *cred, NFSPROC_T *p, int callcnt,
1287     struct nfsclclient *clp, void *id, int flags,
1288     struct nfscllockowner **lpp, int *dorpcp)
1289 {
1290 	struct nfscllockowner *lp;
1291 	struct nfsclopen *op;
1292 	struct nfscllock *nlop, *other_lop = NULL;
1293 	struct nfscldeleg *dp;
1294 	struct nfsnode *np;
1295 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
1296 	int ret = 0, fnd;
1297 
1298 	np = VTONFS(vp);
1299 	*lpp = NULL;
1300 	*dorpcp = 0;
1301 
1302 	/*
1303 	 * Might need these, so MALLOC them now, to
1304 	 * avoid a tsleep() in MALLOC later.
1305 	 */
1306 	nlop = malloc(
1307 	    sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1308 	nlop->nfslo_type = F_UNLCK;
1309 	nlop->nfslo_first = off;
1310 	if (len == NFS64BITSSET) {
1311 		nlop->nfslo_end = NFS64BITSSET;
1312 	} else {
1313 		nlop->nfslo_end = off + len;
1314 		if (nlop->nfslo_end <= nlop->nfslo_first) {
1315 			free(nlop, M_NFSCLLOCK);
1316 			return (NFSERR_INVAL);
1317 		}
1318 	}
1319 	if (callcnt == 0) {
1320 		other_lop = malloc(
1321 		    sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1322 		*other_lop = *nlop;
1323 	}
1324 	nfscl_filllockowner(id, own, flags);
1325 	dp = NULL;
1326 	NFSLOCKCLSTATE();
1327 	if (callcnt == 0)
1328 		dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1329 		    np->n_fhp->nfh_len);
1330 
1331 	/*
1332 	 * First, unlock any local regions on a delegation.
1333 	 */
1334 	if (dp != NULL) {
1335 		/* Look for this lockowner. */
1336 		LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1337 			if (!NFSBCMP(lp->nfsl_owner, own,
1338 			    NFSV4CL_LOCKNAMELEN))
1339 				break;
1340 		}
1341 		if (lp != NULL)
1342 			/* Use other_lop, so nlop is still available */
1343 			(void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1344 	}
1345 
1346 	/*
1347 	 * Now, find a matching open/lockowner that hasn't already been done,
1348 	 * as marked by nfsl_inprog.
1349 	 */
1350 	lp = NULL;
1351 	fnd = 0;
1352 	LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1353 	    np->n_fhp->nfh_len), nfso_hash) {
1354 		if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1355 		    !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1356 			LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1357 				if (lp->nfsl_inprog == NULL &&
1358 				    !NFSBCMP(lp->nfsl_owner, own,
1359 				     NFSV4CL_LOCKNAMELEN)) {
1360 					fnd = 1;
1361 					break;
1362 				}
1363 			}
1364 		}
1365 		if (fnd)
1366 			break;
1367 	}
1368 
1369 	if (lp != NULL) {
1370 		ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1371 		if (ret)
1372 			*dorpcp = 1;
1373 		/*
1374 		 * Serial modifications on the lock owner for multiple
1375 		 * threads for the same process using a read/write lock.
1376 		 */
1377 		lp->nfsl_inprog = p;
1378 		nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1379 		*lpp = lp;
1380 	}
1381 	NFSUNLOCKCLSTATE();
1382 	if (nlop)
1383 		free(nlop, M_NFSCLLOCK);
1384 	if (other_lop)
1385 		free(other_lop, M_NFSCLLOCK);
1386 	return (0);
1387 }
1388 
1389 /*
1390  * Release all lockowners marked in progess for this process and file.
1391  */
1392 void
1393 nfscl_releasealllocks(struct nfsclclient *clp, vnode_t vp, NFSPROC_T *p,
1394     void *id, int flags)
1395 {
1396 	struct nfsclopen *op;
1397 	struct nfscllockowner *lp;
1398 	struct nfsnode *np;
1399 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
1400 
1401 	np = VTONFS(vp);
1402 	nfscl_filllockowner(id, own, flags);
1403 	NFSLOCKCLSTATE();
1404 	LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1405 	    np->n_fhp->nfh_len), nfso_hash) {
1406 		if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1407 		    !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1408 			LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1409 				if (lp->nfsl_inprog == p &&
1410 				    !NFSBCMP(lp->nfsl_owner, own,
1411 				    NFSV4CL_LOCKNAMELEN)) {
1412 					lp->nfsl_inprog = NULL;
1413 					nfscl_lockunlock(&lp->nfsl_rwlock);
1414 				}
1415 			}
1416 		}
1417 	}
1418 	nfscl_clrelease(clp);
1419 	NFSUNLOCKCLSTATE();
1420 }
1421 
1422 /*
1423  * Called to find out if any bytes within the byte range specified are
1424  * write locked by the calling process. Used to determine if flushing
1425  * is required before a LockU.
1426  * If in doubt, return 1, so the flush will occur.
1427  */
1428 int
1429 nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1430     struct ucred *cred, NFSPROC_T *p, void *id, int flags)
1431 {
1432 	struct nfscllockowner *lp;
1433 	struct nfsclopen *op;
1434 	struct nfsclclient *clp;
1435 	struct nfscllock *lop;
1436 	struct nfscldeleg *dp;
1437 	struct nfsnode *np;
1438 	u_int64_t off, end;
1439 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
1440 	int error = 0;
1441 
1442 	np = VTONFS(vp);
1443 	switch (fl->l_whence) {
1444 	case SEEK_SET:
1445 	case SEEK_CUR:
1446 		/*
1447 		 * Caller is responsible for adding any necessary offset
1448 		 * when SEEK_CUR is used.
1449 		 */
1450 		off = fl->l_start;
1451 		break;
1452 	case SEEK_END:
1453 		off = np->n_size + fl->l_start;
1454 		break;
1455 	default:
1456 		return (1);
1457 	}
1458 	if (fl->l_len != 0) {
1459 		end = off + fl->l_len;
1460 		if (end < off)
1461 			return (1);
1462 	} else {
1463 		end = NFS64BITSSET;
1464 	}
1465 
1466 	error = nfscl_getcl(vp->v_mount, cred, p, false, true, &clp);
1467 	if (error)
1468 		return (1);
1469 	nfscl_filllockowner(id, own, flags);
1470 	NFSLOCKCLSTATE();
1471 
1472 	/*
1473 	 * First check the delegation locks.
1474 	 */
1475 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1476 	if (dp != NULL) {
1477 		/* No need to flush if it is a write delegation. */
1478 		if ((dp->nfsdl_flags & NFSCLDL_WRITE) != 0) {
1479 			nfscl_clrelease(clp);
1480 			NFSUNLOCKCLSTATE();
1481 			return (0);
1482 		}
1483 		LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1484 			if (!NFSBCMP(lp->nfsl_owner, own,
1485 			    NFSV4CL_LOCKNAMELEN))
1486 				break;
1487 		}
1488 		if (lp != NULL) {
1489 			LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1490 				if (lop->nfslo_first >= end)
1491 					break;
1492 				if (lop->nfslo_end <= off)
1493 					continue;
1494 				if (lop->nfslo_type == F_WRLCK) {
1495 					nfscl_clrelease(clp);
1496 					NFSUNLOCKCLSTATE();
1497 					return (1);
1498 				}
1499 			}
1500 		}
1501 	}
1502 
1503 	/*
1504 	 * Now, check state against the server.
1505 	 */
1506 	LIST_FOREACH(op, NFSCLOPENHASH(clp, np->n_fhp->nfh_fh,
1507 	    np->n_fhp->nfh_len), nfso_hash) {
1508 		if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1509 		    !NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1510 			LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1511 				if (!NFSBCMP(lp->nfsl_owner, own,
1512 				    NFSV4CL_LOCKNAMELEN))
1513 					break;
1514 			}
1515 			if (lp != NULL) {
1516 				LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1517 					if (lop->nfslo_first >= end)
1518 						break;
1519 					if (lop->nfslo_end <= off)
1520 						continue;
1521 					if (lop->nfslo_type == F_WRLCK) {
1522 						nfscl_clrelease(clp);
1523 						NFSUNLOCKCLSTATE();
1524 						return (1);
1525 					}
1526 				}
1527 			}
1528 		}
1529 	}
1530 	nfscl_clrelease(clp);
1531 	NFSUNLOCKCLSTATE();
1532 	return (0);
1533 }
1534 
1535 /*
1536  * Release a byte range lock owner structure.
1537  */
1538 void
1539 nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1540 {
1541 	struct nfsclclient *clp;
1542 
1543 	if (lp == NULL)
1544 		return;
1545 	NFSLOCKCLSTATE();
1546 	clp = lp->nfsl_open->nfso_own->nfsow_clp;
1547 	if (error != 0 && candelete &&
1548 	    (lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1549 		nfscl_freelockowner(lp, 0);
1550 	else
1551 		nfscl_lockunlock(&lp->nfsl_rwlock);
1552 	nfscl_clrelease(clp);
1553 	NFSUNLOCKCLSTATE();
1554 }
1555 
1556 /*
1557  * Unlink the open structure.
1558  */
1559 static void
1560 nfscl_unlinkopen(struct nfsclopen *op)
1561 {
1562 
1563 	LIST_REMOVE(op, nfso_list);
1564 	if (op->nfso_hash.le_prev != NULL)
1565 		LIST_REMOVE(op, nfso_hash);
1566 }
1567 
1568 /*
1569  * Free up an open structure and any associated byte range lock structures.
1570  */
1571 void
1572 nfscl_freeopen(struct nfsclopen *op, int local, bool unlink)
1573 {
1574 
1575 	if (unlink)
1576 		nfscl_unlinkopen(op);
1577 	nfscl_freealllocks(&op->nfso_lock, local);
1578 	free(op, M_NFSCLOPEN);
1579 	if (local)
1580 		nfsstatsv1.cllocalopens--;
1581 	else
1582 		nfsstatsv1.clopens--;
1583 }
1584 
1585 /*
1586  * Free up all lock owners and associated locks.
1587  */
1588 static void
1589 nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1590 {
1591 	struct nfscllockowner *lp, *nlp;
1592 
1593 	LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1594 		if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1595 			panic("nfscllckw");
1596 		nfscl_freelockowner(lp, local);
1597 	}
1598 }
1599 
1600 /*
1601  * Called for an Open when NFSERR_EXPIRED is received from the server.
1602  * If there are no byte range locks nor a Share Deny lost, try to do a
1603  * fresh Open. Otherwise, free the open.
1604  */
1605 static int
1606 nfscl_expireopen(struct nfsclclient *clp, struct nfsclopen *op,
1607     struct nfsmount *nmp, struct ucred *cred, NFSPROC_T *p)
1608 {
1609 	struct nfscllockowner *lp;
1610 	struct nfscldeleg *dp;
1611 	int mustdelete = 0, error;
1612 
1613 	/*
1614 	 * Look for any byte range lock(s).
1615 	 */
1616 	LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1617 		if (!LIST_EMPTY(&lp->nfsl_lock)) {
1618 			mustdelete = 1;
1619 			break;
1620 		}
1621 	}
1622 
1623 	/*
1624 	 * If no byte range lock(s) nor a Share deny, try to re-open.
1625 	 */
1626 	if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1627 		newnfs_copycred(&op->nfso_cred, cred);
1628 		dp = NULL;
1629 		error = nfsrpc_reopen(nmp, op->nfso_fh,
1630 		    op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1631 		if (error) {
1632 			mustdelete = 1;
1633 			if (dp != NULL) {
1634 				free(dp, M_NFSCLDELEG);
1635 				dp = NULL;
1636 			}
1637 		}
1638 		if (dp != NULL)
1639 			nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1640 			    op->nfso_fhlen, cred, p, &dp);
1641 	}
1642 
1643 	/*
1644 	 * If a byte range lock or Share deny or couldn't re-open, free it.
1645 	 */
1646 	if (mustdelete)
1647 		nfscl_freeopen(op, 0, true);
1648 	return (mustdelete);
1649 }
1650 
1651 /*
1652  * Free up an open owner structure.
1653  */
1654 static void
1655 nfscl_freeopenowner(struct nfsclowner *owp, int local)
1656 {
1657 	int owned;
1658 
1659 	/*
1660 	 * Make sure the NFSCLSTATE mutex is held, to avoid races with
1661 	 * calls in nfscl_renewthread() that do not hold a reference
1662 	 * count on the nfsclclient and just the mutex.
1663 	 * The mutex will not be held for calls done with the exclusive
1664 	 * nfsclclient lock held, in particular, nfscl_hasexpired()
1665 	 * and nfscl_recalldeleg() might do this.
1666 	 */
1667 	owned = mtx_owned(NFSCLSTATEMUTEXPTR);
1668 	if (owned == 0)
1669 		NFSLOCKCLSTATE();
1670 	LIST_REMOVE(owp, nfsow_list);
1671 	if (owned == 0)
1672 		NFSUNLOCKCLSTATE();
1673 	free(owp, M_NFSCLOWNER);
1674 	if (local)
1675 		nfsstatsv1.cllocalopenowners--;
1676 	else
1677 		nfsstatsv1.clopenowners--;
1678 }
1679 
1680 /*
1681  * Free up a byte range lock owner structure.
1682  */
1683 void
1684 nfscl_freelockowner(struct nfscllockowner *lp, int local)
1685 {
1686 	struct nfscllock *lop, *nlop;
1687 	int owned;
1688 
1689 	/*
1690 	 * Make sure the NFSCLSTATE mutex is held, to avoid races with
1691 	 * calls in nfscl_renewthread() that do not hold a reference
1692 	 * count on the nfsclclient and just the mutex.
1693 	 * The mutex will not be held for calls done with the exclusive
1694 	 * nfsclclient lock held, in particular, nfscl_hasexpired()
1695 	 * and nfscl_recalldeleg() might do this.
1696 	 */
1697 	owned = mtx_owned(NFSCLSTATEMUTEXPTR);
1698 	if (owned == 0)
1699 		NFSLOCKCLSTATE();
1700 	LIST_REMOVE(lp, nfsl_list);
1701 	if (owned == 0)
1702 		NFSUNLOCKCLSTATE();
1703 	LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1704 		nfscl_freelock(lop, local);
1705 	}
1706 	free(lp, M_NFSCLLOCKOWNER);
1707 	if (local)
1708 		nfsstatsv1.cllocallockowners--;
1709 	else
1710 		nfsstatsv1.cllockowners--;
1711 }
1712 
1713 /*
1714  * Free up a byte range lock structure.
1715  */
1716 void
1717 nfscl_freelock(struct nfscllock *lop, int local)
1718 {
1719 
1720 	LIST_REMOVE(lop, nfslo_list);
1721 	free(lop, M_NFSCLLOCK);
1722 	if (local)
1723 		nfsstatsv1.cllocallocks--;
1724 	else
1725 		nfsstatsv1.cllocks--;
1726 }
1727 
1728 /*
1729  * Clean out the state related to a delegation.
1730  */
1731 static void
1732 nfscl_cleandeleg(struct nfscldeleg *dp)
1733 {
1734 	struct nfsclowner *owp, *nowp;
1735 	struct nfsclopen *op;
1736 
1737 	LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1738 		op = LIST_FIRST(&owp->nfsow_open);
1739 		if (op != NULL) {
1740 			if (LIST_NEXT(op, nfso_list) != NULL)
1741 				panic("nfscleandel");
1742 			nfscl_freeopen(op, 1, true);
1743 		}
1744 		nfscl_freeopenowner(owp, 1);
1745 	}
1746 	nfscl_freealllocks(&dp->nfsdl_lock, 1);
1747 }
1748 
1749 /*
1750  * Free a delegation.
1751  */
1752 static void
1753 nfscl_freedeleg(struct nfscldeleghead *hdp, struct nfscldeleg *dp, bool freeit)
1754 {
1755 
1756 	TAILQ_REMOVE(hdp, dp, nfsdl_list);
1757 	LIST_REMOVE(dp, nfsdl_hash);
1758 	if (freeit)
1759 		free(dp, M_NFSCLDELEG);
1760 	nfsstatsv1.cldelegates--;
1761 	nfscl_delegcnt--;
1762 }
1763 
1764 /*
1765  * Free up all state related to this client structure.
1766  */
1767 static void
1768 nfscl_cleanclient(struct nfsclclient *clp)
1769 {
1770 	struct nfsclowner *owp, *nowp;
1771 	struct nfsclopen *op, *nop;
1772 	struct nfscllayout *lyp, *nlyp;
1773 	struct nfscldevinfo *dip, *ndip;
1774 
1775 	TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1776 		nfscl_freelayout(lyp);
1777 
1778 	LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1779 		nfscl_freedevinfo(dip);
1780 
1781 	/* Now, all the OpenOwners, etc. */
1782 	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1783 		LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1784 			nfscl_freeopen(op, 0, true);
1785 		}
1786 		nfscl_freeopenowner(owp, 0);
1787 	}
1788 }
1789 
1790 /*
1791  * Called when an NFSERR_EXPIRED is received from the server.
1792  */
1793 static void
1794 nfscl_expireclient(struct nfsclclient *clp, struct nfsmount *nmp,
1795     struct ucred *cred, NFSPROC_T *p)
1796 {
1797 	struct nfsclowner *owp, *nowp, *towp;
1798 	struct nfsclopen *op, *nop, *top;
1799 	struct nfscldeleg *dp, *ndp;
1800 	int ret, printed = 0;
1801 
1802 	/*
1803 	 * First, merge locally issued Opens into the list for the server.
1804 	 */
1805 	dp = TAILQ_FIRST(&clp->nfsc_deleg);
1806 	while (dp != NULL) {
1807 	    ndp = TAILQ_NEXT(dp, nfsdl_list);
1808 	    owp = LIST_FIRST(&dp->nfsdl_owner);
1809 	    while (owp != NULL) {
1810 		nowp = LIST_NEXT(owp, nfsow_list);
1811 		op = LIST_FIRST(&owp->nfsow_open);
1812 		if (op != NULL) {
1813 		    if (LIST_NEXT(op, nfso_list) != NULL)
1814 			panic("nfsclexp");
1815 		    LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1816 			if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1817 			    NFSV4CL_LOCKNAMELEN))
1818 			    break;
1819 		    }
1820 		    if (towp != NULL) {
1821 			/* Merge opens in */
1822 			LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1823 			    if (top->nfso_fhlen == op->nfso_fhlen &&
1824 				!NFSBCMP(top->nfso_fh, op->nfso_fh,
1825 				 op->nfso_fhlen)) {
1826 				top->nfso_mode |= op->nfso_mode;
1827 				top->nfso_opencnt += op->nfso_opencnt;
1828 				break;
1829 			    }
1830 			}
1831 			if (top == NULL) {
1832 			    /* Just add the open to the owner list */
1833 			    LIST_REMOVE(op, nfso_list);
1834 			    op->nfso_own = towp;
1835 			    LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1836 			    LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
1837 				op->nfso_fhlen), op, nfso_hash);
1838 			    nfsstatsv1.cllocalopens--;
1839 			    nfsstatsv1.clopens++;
1840 			}
1841 		    } else {
1842 			/* Just add the openowner to the client list */
1843 			LIST_REMOVE(owp, nfsow_list);
1844 			owp->nfsow_clp = clp;
1845 			LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1846 			LIST_INSERT_HEAD(NFSCLOPENHASH(clp, op->nfso_fh,
1847 			    op->nfso_fhlen), op, nfso_hash);
1848 			nfsstatsv1.cllocalopenowners--;
1849 			nfsstatsv1.clopenowners++;
1850 			nfsstatsv1.cllocalopens--;
1851 			nfsstatsv1.clopens++;
1852 		    }
1853 		}
1854 		owp = nowp;
1855 	    }
1856 	    if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1857 		printed = 1;
1858 		printf("nfsv4 expired locks lost\n");
1859 	    }
1860 	    nfscl_cleandeleg(dp);
1861 	    nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
1862 	    dp = ndp;
1863 	}
1864 	if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1865 	    panic("nfsclexp");
1866 
1867 	/*
1868 	 * Now, try and reopen against the server.
1869 	 */
1870 	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1871 		owp->nfsow_seqid = 0;
1872 		LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1873 			ret = nfscl_expireopen(clp, op, nmp, cred, p);
1874 			if (ret && !printed) {
1875 				printed = 1;
1876 				printf("nfsv4 expired locks lost\n");
1877 			}
1878 		}
1879 		if (LIST_EMPTY(&owp->nfsow_open))
1880 			nfscl_freeopenowner(owp, 0);
1881 	}
1882 }
1883 
1884 /*
1885  * This function must be called after the process represented by "own" has
1886  * exited. Must be called with CLSTATE lock held.
1887  */
1888 static void
1889 nfscl_cleanup_common(struct nfsclclient *clp, u_int8_t *own)
1890 {
1891 	struct nfsclowner *owp, *nowp;
1892 	struct nfscllockowner *lp;
1893 	struct nfscldeleg *dp;
1894 
1895 	/* First, get rid of local locks on delegations. */
1896 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1897 		LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1898 		    if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1899 			if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1900 			    panic("nfscllckw");
1901 			nfscl_freelockowner(lp, 1);
1902 			break;
1903 		    }
1904 		}
1905 	}
1906 	owp = LIST_FIRST(&clp->nfsc_owner);
1907 	while (owp != NULL) {
1908 		nowp = LIST_NEXT(owp, nfsow_list);
1909 		if (!NFSBCMP(owp->nfsow_owner, own,
1910 		    NFSV4CL_LOCKNAMELEN)) {
1911 			/*
1912 			 * If there are children that haven't closed the
1913 			 * file descriptors yet, the opens will still be
1914 			 * here. For that case, let the renew thread clear
1915 			 * out the OpenOwner later.
1916 			 */
1917 			if (LIST_EMPTY(&owp->nfsow_open))
1918 				nfscl_freeopenowner(owp, 0);
1919 			else
1920 				owp->nfsow_defunct = 1;
1921 			break;
1922 		}
1923 		owp = nowp;
1924 	}
1925 }
1926 
1927 /*
1928  * Find open/lock owners for processes that have exited.
1929  */
1930 static void
1931 nfscl_cleanupkext(struct nfsclclient *clp, struct nfscllockownerfhhead *lhp)
1932 {
1933 	struct nfsclowner *owp, *nowp;
1934 	struct nfsclopen *op;
1935 	struct nfscllockowner *lp, *nlp;
1936 	struct nfscldeleg *dp;
1937 	uint8_t own[NFSV4CL_LOCKNAMELEN];
1938 
1939 	/*
1940 	 * All the pidhash locks must be acquired, since they are sx locks
1941 	 * and must be acquired before the mutexes.  The pid(s) that will
1942 	 * be used aren't known yet, so all the locks need to be acquired.
1943 	 * Fortunately, this function is only performed once/sec.
1944 	 */
1945 	pidhash_slockall();
1946 	NFSLOCKCLSTATE();
1947 	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1948 		LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1949 			LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1950 				if (LIST_EMPTY(&lp->nfsl_lock))
1951 					nfscl_emptylockowner(lp, lhp);
1952 			}
1953 		}
1954 		if (nfscl_procdoesntexist(owp->nfsow_owner)) {
1955 			memcpy(own, owp->nfsow_owner, NFSV4CL_LOCKNAMELEN);
1956 			nfscl_cleanup_common(clp, own);
1957 		}
1958 	}
1959 
1960 	/*
1961 	 * For the single open_owner case, these lock owners need to be
1962 	 * checked to see if they still exist separately.
1963 	 * This is because nfscl_procdoesntexist() never returns true for
1964 	 * the single open_owner so that the above doesn't ever call
1965 	 * nfscl_cleanup_common().
1966 	 */
1967 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1968 		LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1969 			if (nfscl_procdoesntexist(lp->nfsl_owner)) {
1970 				memcpy(own, lp->nfsl_owner,
1971 				    NFSV4CL_LOCKNAMELEN);
1972 				nfscl_cleanup_common(clp, own);
1973 			}
1974 		}
1975 	}
1976 	NFSUNLOCKCLSTATE();
1977 	pidhash_sunlockall();
1978 }
1979 
1980 /*
1981  * Take the empty lock owner and move it to the local lhp list if the
1982  * associated process no longer exists.
1983  */
1984 static void
1985 nfscl_emptylockowner(struct nfscllockowner *lp,
1986     struct nfscllockownerfhhead *lhp)
1987 {
1988 	struct nfscllockownerfh *lfhp, *mylfhp;
1989 	struct nfscllockowner *nlp;
1990 	int fnd_it;
1991 
1992 	/* If not a Posix lock owner, just return. */
1993 	if ((lp->nfsl_lockflags & F_POSIX) == 0)
1994 		return;
1995 
1996 	fnd_it = 0;
1997 	mylfhp = NULL;
1998 	/*
1999 	 * First, search to see if this lock owner is already in the list.
2000 	 * If it is, then the associated process no longer exists.
2001 	 */
2002 	SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
2003 		if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
2004 		    !NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
2005 		    lfhp->nfslfh_len))
2006 			mylfhp = lfhp;
2007 		LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
2008 			if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
2009 			    NFSV4CL_LOCKNAMELEN))
2010 				fnd_it = 1;
2011 	}
2012 	/* If not found, check if process still exists. */
2013 	if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
2014 		return;
2015 
2016 	/* Move the lock owner over to the local list. */
2017 	if (mylfhp == NULL) {
2018 		mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
2019 		    M_NOWAIT);
2020 		if (mylfhp == NULL)
2021 			return;
2022 		mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
2023 		NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
2024 		    mylfhp->nfslfh_len);
2025 		LIST_INIT(&mylfhp->nfslfh_lock);
2026 		SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
2027 	}
2028 	LIST_REMOVE(lp, nfsl_list);
2029 	LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
2030 }
2031 
2032 static int	fake_global;	/* Used to force visibility of MNTK_UNMOUNTF */
2033 /*
2034  * Called from nfs umount to free up the clientid.
2035  */
2036 void
2037 nfscl_umount(struct nfsmount *nmp, NFSPROC_T *p, struct nfscldeleghead *dhp)
2038 {
2039 	struct nfsclclient *clp;
2040 	struct ucred *cred;
2041 	int igotlock;
2042 
2043 	/*
2044 	 * For the case that matters, this is the thread that set
2045 	 * MNTK_UNMOUNTF, so it will see it set. The code that follows is
2046 	 * done to ensure that any thread executing nfscl_getcl() after
2047 	 * this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
2048 	 * mutex for NFSLOCKCLSTATE(), so it is "m" for the following
2049 	 * explanation, courtesy of Alan Cox.
2050 	 * What follows is a snippet from Alan Cox's email at:
2051 	 * https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
2052 	 *
2053 	 * 1. Set MNTK_UNMOUNTF
2054 	 * 2. Acquire a standard FreeBSD mutex "m".
2055 	 * 3. Update some data structures.
2056 	 * 4. Release mutex "m".
2057 	 *
2058 	 * Then, other threads that acquire "m" after step 4 has occurred will
2059 	 * see MNTK_UNMOUNTF as set.  But, other threads that beat thread X to
2060 	 * step 2 may or may not see MNTK_UNMOUNTF as set.
2061 	 */
2062 	NFSLOCKCLSTATE();
2063 	if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
2064 		fake_global++;
2065 		NFSUNLOCKCLSTATE();
2066 		NFSLOCKCLSTATE();
2067 	}
2068 
2069 	clp = nmp->nm_clp;
2070 	if (clp != NULL) {
2071 		if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
2072 			panic("nfscl umount");
2073 
2074 		/*
2075 		 * First, handshake with the nfscl renew thread, to terminate
2076 		 * it.
2077 		 */
2078 		clp->nfsc_flags |= NFSCLFLAGS_UMOUNT;
2079 		while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
2080 			(void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
2081 			    "nfsclumnt", hz);
2082 
2083 		/*
2084 		 * Now, get the exclusive lock on the client state, so
2085 		 * that no uses of the state are still in progress.
2086 		 */
2087 		do {
2088 			igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2089 			    NFSCLSTATEMUTEXPTR, NULL);
2090 		} while (!igotlock);
2091 		NFSUNLOCKCLSTATE();
2092 
2093 		/*
2094 		 * Free up all the state. It will expire on the server, but
2095 		 * maybe we should do a SetClientId/SetClientIdConfirm so
2096 		 * the server throws it away?
2097 		 */
2098 		LIST_REMOVE(clp, nfsc_list);
2099 		nfscl_delegreturnall(clp, p, dhp);
2100 		cred = newnfs_getcred();
2101 		if (NFSHASNFSV4N(nmp)) {
2102 			nfsrpc_destroysession(nmp, NULL, cred, p);
2103 			nfsrpc_destroyclient(nmp, clp, cred, p);
2104 		} else
2105 			nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2106 		nfscl_cleanclient(clp);
2107 		nmp->nm_clp = NULL;
2108 		NFSFREECRED(cred);
2109 		free(clp, M_NFSCLCLIENT);
2110 	} else
2111 		NFSUNLOCKCLSTATE();
2112 }
2113 
2114 /*
2115  * This function is called when a server replies with NFSERR_STALECLIENTID
2116  * NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
2117  * doing Opens and Locks with reclaim. If these fail, it deletes the
2118  * corresponding state.
2119  */
2120 static void
2121 nfscl_recover(struct nfsclclient *clp, bool *retokp, struct ucred *cred,
2122     NFSPROC_T *p)
2123 {
2124 	struct nfsclowner *owp, *nowp;
2125 	struct nfsclopen *op, *nop;
2126 	struct nfscllockowner *lp, *nlp;
2127 	struct nfscllock *lop, *nlop;
2128 	struct nfscldeleg *dp, *ndp, *tdp;
2129 	struct nfsmount *nmp;
2130 	struct ucred *tcred;
2131 	struct nfsclopenhead extra_open;
2132 	struct nfscldeleghead extra_deleg;
2133 	struct nfsreq *rep;
2134 	u_int64_t len;
2135 	u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
2136 	int i, igotlock = 0, error, trycnt, firstlock;
2137 	struct nfscllayout *lyp, *nlyp;
2138 	bool recovered_one;
2139 
2140 	/*
2141 	 * First, lock the client structure, so everyone else will
2142 	 * block when trying to use state.
2143 	 */
2144 	NFSLOCKCLSTATE();
2145 	clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2146 	do {
2147 		igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2148 		    NFSCLSTATEMUTEXPTR, NULL);
2149 	} while (!igotlock);
2150 	NFSUNLOCKCLSTATE();
2151 
2152 	nmp = clp->nfsc_nmp;
2153 	if (nmp == NULL)
2154 		panic("nfscl recover");
2155 
2156 	/*
2157 	 * For now, just get rid of all layouts. There may be a need
2158 	 * to do LayoutCommit Ops with reclaim == true later.
2159 	 */
2160 	TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2161 		nfscl_freelayout(lyp);
2162 	TAILQ_INIT(&clp->nfsc_layout);
2163 	for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2164 		LIST_INIT(&clp->nfsc_layouthash[i]);
2165 
2166 	trycnt = 5;
2167 	tcred = NULL;
2168 	do {
2169 		error = nfsrpc_setclient(nmp, clp, 1, retokp, cred, p);
2170 	} while ((error == NFSERR_STALECLIENTID ||
2171 	     error == NFSERR_BADSESSION ||
2172 	     error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2173 	if (error) {
2174 		NFSLOCKCLSTATE();
2175 		clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER |
2176 		    NFSCLFLAGS_RECVRINPROG);
2177 		wakeup(&clp->nfsc_flags);
2178 		nfsv4_unlock(&clp->nfsc_lock, 0);
2179 		NFSUNLOCKCLSTATE();
2180 		return;
2181 	}
2182 	clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2183 	clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2184 
2185 	/*
2186 	 * Mark requests already queued on the server, so that they don't
2187 	 * initiate another recovery cycle. Any requests already in the
2188 	 * queue that handle state information will have the old stale
2189 	 * clientid/stateid and will get a NFSERR_STALESTATEID,
2190 	 * NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2191 	 * This will be translated to NFSERR_STALEDONTRECOVER when
2192 	 * R_DONTRECOVER is set.
2193 	 */
2194 	NFSLOCKREQ();
2195 	TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2196 		if (rep->r_nmp == nmp)
2197 			rep->r_flags |= R_DONTRECOVER;
2198 	}
2199 	NFSUNLOCKREQ();
2200 
2201 	/*
2202 	 * If nfsrpc_setclient() returns *retokp == true,
2203 	 * no more recovery is needed.
2204 	 */
2205 	if (*retokp)
2206 		goto out;
2207 
2208 	/*
2209 	 * Now, mark all delegations "need reclaim".
2210 	 */
2211 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2212 		dp->nfsdl_flags |= NFSCLDL_NEEDRECLAIM;
2213 
2214 	TAILQ_INIT(&extra_deleg);
2215 	LIST_INIT(&extra_open);
2216 	/*
2217 	 * Now traverse the state lists, doing Open and Lock Reclaims.
2218 	 */
2219 	tcred = newnfs_getcred();
2220 	recovered_one = false;
2221 	owp = LIST_FIRST(&clp->nfsc_owner);
2222 	while (owp != NULL) {
2223 	    nowp = LIST_NEXT(owp, nfsow_list);
2224 	    owp->nfsow_seqid = 0;
2225 	    op = LIST_FIRST(&owp->nfsow_open);
2226 	    while (op != NULL) {
2227 		nop = LIST_NEXT(op, nfso_list);
2228 		if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2229 		    /* Search for a delegation to reclaim with the open */
2230 		    TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2231 			if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2232 			    continue;
2233 			if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2234 			    mode = NFSV4OPEN_ACCESSWRITE;
2235 			    delegtype = NFSV4OPEN_DELEGATEWRITE;
2236 			} else {
2237 			    mode = NFSV4OPEN_ACCESSREAD;
2238 			    delegtype = NFSV4OPEN_DELEGATEREAD;
2239 			}
2240 			if ((op->nfso_mode & mode) == mode &&
2241 			    op->nfso_fhlen == dp->nfsdl_fhlen &&
2242 			    !NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2243 			    break;
2244 		    }
2245 		    ndp = dp;
2246 		    if (dp == NULL)
2247 			delegtype = NFSV4OPEN_DELEGATENONE;
2248 		    newnfs_copycred(&op->nfso_cred, tcred);
2249 		    error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2250 			op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2251 			op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2252 			tcred, p);
2253 		    if (!error) {
2254 			recovered_one = true;
2255 			/* Handle any replied delegation */
2256 			if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2257 			    || NFSMNT_RDONLY(nmp->nm_mountp))) {
2258 			    if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2259 				mode = NFSV4OPEN_ACCESSWRITE;
2260 			    else
2261 				mode = NFSV4OPEN_ACCESSREAD;
2262 			    TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2263 				if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2264 				    continue;
2265 				if ((op->nfso_mode & mode) == mode &&
2266 				    op->nfso_fhlen == dp->nfsdl_fhlen &&
2267 				    !NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2268 				    op->nfso_fhlen)) {
2269 				    dp->nfsdl_stateid = ndp->nfsdl_stateid;
2270 				    dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2271 				    dp->nfsdl_ace = ndp->nfsdl_ace;
2272 				    dp->nfsdl_change = ndp->nfsdl_change;
2273 				    dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2274 				    if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2275 					dp->nfsdl_flags |= NFSCLDL_RECALL;
2276 				    free(ndp, M_NFSCLDELEG);
2277 				    ndp = NULL;
2278 				    break;
2279 				}
2280 			    }
2281 			}
2282 			if (ndp != NULL)
2283 			    TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2284 
2285 			/* and reclaim all byte range locks */
2286 			lp = LIST_FIRST(&op->nfso_lock);
2287 			while (lp != NULL) {
2288 			    nlp = LIST_NEXT(lp, nfsl_list);
2289 			    lp->nfsl_seqid = 0;
2290 			    firstlock = 1;
2291 			    lop = LIST_FIRST(&lp->nfsl_lock);
2292 			    while (lop != NULL) {
2293 				nlop = LIST_NEXT(lop, nfslo_list);
2294 				if (lop->nfslo_end == NFS64BITSSET)
2295 				    len = NFS64BITSSET;
2296 				else
2297 				    len = lop->nfslo_end - lop->nfslo_first;
2298 				error = nfscl_trylock(nmp, NULL,
2299 				    op->nfso_fh, op->nfso_fhlen, lp,
2300 				    firstlock, 1, lop->nfslo_first, len,
2301 				    lop->nfslo_type, tcred, p);
2302 				if (error != 0)
2303 				    nfscl_freelock(lop, 0);
2304 				else
2305 				    firstlock = 0;
2306 				lop = nlop;
2307 			    }
2308 			    /* If no locks, but a lockowner, just delete it. */
2309 			    if (LIST_EMPTY(&lp->nfsl_lock))
2310 				nfscl_freelockowner(lp, 0);
2311 			    lp = nlp;
2312 			}
2313 		    } else if (error == NFSERR_NOGRACE && !recovered_one &&
2314 			NFSHASNFSV4N(nmp)) {
2315 			/*
2316 			 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2317 			 * actually end up here, since the client will do
2318 			 * a recovery for NFSERR_BADSESSION, but will get
2319 			 * an NFSERR_NOGRACE reply for the first "reclaim"
2320 			 * attempt.
2321 			 * So, call nfscl_expireclient() to recover the
2322 			 * opens as best we can and then do a reclaim
2323 			 * complete and return.
2324 			 */
2325 			nfsrpc_reclaimcomplete(nmp, cred, p);
2326 			nfscl_expireclient(clp, nmp, tcred, p);
2327 			goto out;
2328 		    }
2329 		}
2330 		if (error != 0 && error != NFSERR_BADSESSION)
2331 		    nfscl_freeopen(op, 0, true);
2332 		op = nop;
2333 	    }
2334 	    owp = nowp;
2335 	}
2336 
2337 	/*
2338 	 * Now, try and get any delegations not yet reclaimed by cobbling
2339 	 * to-gether an appropriate open.
2340 	 */
2341 	nowp = NULL;
2342 	dp = TAILQ_FIRST(&clp->nfsc_deleg);
2343 	while (dp != NULL) {
2344 	    ndp = TAILQ_NEXT(dp, nfsdl_list);
2345 	    if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2346 		if (nowp == NULL) {
2347 		    nowp = malloc(
2348 			sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2349 		    /*
2350 		     * Name must be as long an largest possible
2351 		     * NFSV4CL_LOCKNAMELEN. 12 for now.
2352 		     */
2353 		    NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2354 			NFSV4CL_LOCKNAMELEN);
2355 		    LIST_INIT(&nowp->nfsow_open);
2356 		    nowp->nfsow_clp = clp;
2357 		    nowp->nfsow_seqid = 0;
2358 		    nowp->nfsow_defunct = 0;
2359 		    nfscl_lockinit(&nowp->nfsow_rwlock);
2360 		}
2361 		nop = NULL;
2362 		if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2363 		    nop = malloc(sizeof (struct nfsclopen) +
2364 			dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2365 		    nop->nfso_own = nowp;
2366 		    if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2367 			nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2368 			delegtype = NFSV4OPEN_DELEGATEWRITE;
2369 		    } else {
2370 			nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2371 			delegtype = NFSV4OPEN_DELEGATEREAD;
2372 		    }
2373 		    nop->nfso_opencnt = 0;
2374 		    nop->nfso_posixlock = 1;
2375 		    nop->nfso_fhlen = dp->nfsdl_fhlen;
2376 		    NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2377 		    LIST_INIT(&nop->nfso_lock);
2378 		    nop->nfso_stateid.seqid = 0;
2379 		    nop->nfso_stateid.other[0] = 0;
2380 		    nop->nfso_stateid.other[1] = 0;
2381 		    nop->nfso_stateid.other[2] = 0;
2382 		    newnfs_copycred(&dp->nfsdl_cred, tcred);
2383 		    newnfs_copyincred(tcred, &nop->nfso_cred);
2384 		    tdp = NULL;
2385 		    error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2386 			nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2387 			nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2388 			delegtype, tcred, p);
2389 		    if (tdp != NULL) {
2390 			if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2391 			    mode = NFSV4OPEN_ACCESSWRITE;
2392 			else
2393 			    mode = NFSV4OPEN_ACCESSREAD;
2394 			if ((nop->nfso_mode & mode) == mode &&
2395 			    nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2396 			    !NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2397 			    nop->nfso_fhlen)) {
2398 			    dp->nfsdl_stateid = tdp->nfsdl_stateid;
2399 			    dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2400 			    dp->nfsdl_ace = tdp->nfsdl_ace;
2401 			    dp->nfsdl_change = tdp->nfsdl_change;
2402 			    dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2403 			    if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2404 				dp->nfsdl_flags |= NFSCLDL_RECALL;
2405 			    free(tdp, M_NFSCLDELEG);
2406 			} else {
2407 			    TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2408 			}
2409 		    }
2410 		}
2411 		if (error) {
2412 		    if (nop != NULL)
2413 			free(nop, M_NFSCLOPEN);
2414 		    if (error == NFSERR_NOGRACE && !recovered_one &&
2415 			NFSHASNFSV4N(nmp)) {
2416 			/*
2417 			 * For NFSv4.1/4.2, the NFSERR_EXPIRED case will
2418 			 * actually end up here, since the client will do
2419 			 * a recovery for NFSERR_BADSESSION, but will get
2420 			 * an NFSERR_NOGRACE reply for the first "reclaim"
2421 			 * attempt.
2422 			 * So, call nfscl_expireclient() to recover the
2423 			 * opens as best we can and then do a reclaim
2424 			 * complete and return.
2425 			 */
2426 			nfsrpc_reclaimcomplete(nmp, cred, p);
2427 			nfscl_expireclient(clp, nmp, tcred, p);
2428 			free(nowp, M_NFSCLOWNER);
2429 			goto out;
2430 		    }
2431 		    /*
2432 		     * Couldn't reclaim it, so throw the state
2433 		     * away. Ouch!!
2434 		     */
2435 		    nfscl_cleandeleg(dp);
2436 		    nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
2437 		} else {
2438 		    recovered_one = true;
2439 		    LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2440 		}
2441 	    }
2442 	    dp = ndp;
2443 	}
2444 
2445 	/*
2446 	 * Now, get rid of extra Opens and Delegations.
2447 	 */
2448 	LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2449 		do {
2450 			newnfs_copycred(&op->nfso_cred, tcred);
2451 			error = nfscl_tryclose(op, tcred, nmp, p, true);
2452 			if (error == NFSERR_GRACE)
2453 				(void) nfs_catnap(PZERO, error, "nfsexcls");
2454 		} while (error == NFSERR_GRACE);
2455 		LIST_REMOVE(op, nfso_list);
2456 		free(op, M_NFSCLOPEN);
2457 	}
2458 	if (nowp != NULL)
2459 		free(nowp, M_NFSCLOWNER);
2460 
2461 	TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2462 		do {
2463 			newnfs_copycred(&dp->nfsdl_cred, tcred);
2464 			error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2465 			if (error == NFSERR_GRACE)
2466 				(void) nfs_catnap(PZERO, error, "nfsexdlg");
2467 		} while (error == NFSERR_GRACE);
2468 		TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2469 		free(dp, M_NFSCLDELEG);
2470 	}
2471 
2472 	/* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2473 	if (NFSHASNFSV4N(nmp))
2474 		(void)nfsrpc_reclaimcomplete(nmp, cred, p);
2475 
2476 out:
2477 	NFSLOCKCLSTATE();
2478 	clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2479 	wakeup(&clp->nfsc_flags);
2480 	nfsv4_unlock(&clp->nfsc_lock, 0);
2481 	NFSUNLOCKCLSTATE();
2482 	if (tcred != NULL)
2483 		NFSFREECRED(tcred);
2484 }
2485 
2486 /*
2487  * This function is called when a server replies with NFSERR_EXPIRED.
2488  * It deletes all state for the client and does a fresh SetClientId/confirm.
2489  * XXX Someday it should post a signal to the process(es) that hold the
2490  * state, so they know that lock state has been lost.
2491  */
2492 int
2493 nfscl_hasexpired(struct nfsclclient *clp, u_int32_t clidrev, NFSPROC_T *p)
2494 {
2495 	struct nfsmount *nmp;
2496 	struct ucred *cred;
2497 	int igotlock = 0, error, trycnt;
2498 
2499 	/*
2500 	 * If the clientid has gone away or a new SetClientid has already
2501 	 * been done, just return ok.
2502 	 */
2503 	if (clp == NULL || clidrev != clp->nfsc_clientidrev)
2504 		return (0);
2505 
2506 	/*
2507 	 * First, lock the client structure, so everyone else will
2508 	 * block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2509 	 * that only one thread does the work.
2510 	 */
2511 	NFSLOCKCLSTATE();
2512 	clp->nfsc_flags |= NFSCLFLAGS_EXPIREIT;
2513 	do {
2514 		igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2515 		    NFSCLSTATEMUTEXPTR, NULL);
2516 	} while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2517 	if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2518 		if (igotlock)
2519 			nfsv4_unlock(&clp->nfsc_lock, 0);
2520 		NFSUNLOCKCLSTATE();
2521 		return (0);
2522 	}
2523 	clp->nfsc_flags |= NFSCLFLAGS_RECVRINPROG;
2524 	NFSUNLOCKCLSTATE();
2525 
2526 	nmp = clp->nfsc_nmp;
2527 	if (nmp == NULL)
2528 		panic("nfscl expired");
2529 	cred = newnfs_getcred();
2530 	trycnt = 5;
2531 	do {
2532 		error = nfsrpc_setclient(nmp, clp, 0, NULL, cred, p);
2533 	} while ((error == NFSERR_STALECLIENTID ||
2534 	     error == NFSERR_BADSESSION ||
2535 	     error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2536 	if (error) {
2537 		NFSLOCKCLSTATE();
2538 		clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2539 	} else {
2540 		/*
2541 		 * Expire the state for the client.
2542 		 */
2543 		nfscl_expireclient(clp, nmp, cred, p);
2544 		NFSLOCKCLSTATE();
2545 		clp->nfsc_flags |= NFSCLFLAGS_HASCLIENTID;
2546 		clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2547 	}
2548 	clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT | NFSCLFLAGS_RECVRINPROG);
2549 	wakeup(&clp->nfsc_flags);
2550 	nfsv4_unlock(&clp->nfsc_lock, 0);
2551 	NFSUNLOCKCLSTATE();
2552 	NFSFREECRED(cred);
2553 	return (error);
2554 }
2555 
2556 /*
2557  * This function inserts a lock in the list after insert_lop.
2558  */
2559 static void
2560 nfscl_insertlock(struct nfscllockowner *lp, struct nfscllock *new_lop,
2561     struct nfscllock *insert_lop, int local)
2562 {
2563 
2564 	if ((struct nfscllockowner *)insert_lop == lp)
2565 		LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2566 	else
2567 		LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2568 	if (local)
2569 		nfsstatsv1.cllocallocks++;
2570 	else
2571 		nfsstatsv1.cllocks++;
2572 }
2573 
2574 /*
2575  * This function updates the locking for a lock owner and given file. It
2576  * maintains a list of lock ranges ordered on increasing file offset that
2577  * are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2578  * It always adds new_lop to the list and sometimes uses the one pointed
2579  * at by other_lopp.
2580  * Returns 1 if the locks were modified, 0 otherwise.
2581  */
2582 static int
2583 nfscl_updatelock(struct nfscllockowner *lp, struct nfscllock **new_lopp,
2584     struct nfscllock **other_lopp, int local)
2585 {
2586 	struct nfscllock *new_lop = *new_lopp;
2587 	struct nfscllock *lop, *tlop, *ilop;
2588 	struct nfscllock *other_lop;
2589 	int unlock = 0, modified = 0;
2590 	u_int64_t tmp;
2591 
2592 	/*
2593 	 * Work down the list until the lock is merged.
2594 	 */
2595 	if (new_lop->nfslo_type == F_UNLCK)
2596 		unlock = 1;
2597 	ilop = (struct nfscllock *)lp;
2598 	lop = LIST_FIRST(&lp->nfsl_lock);
2599 	while (lop != NULL) {
2600 	    /*
2601 	     * Only check locks for this file that aren't before the start of
2602 	     * new lock's range.
2603 	     */
2604 	    if (lop->nfslo_end >= new_lop->nfslo_first) {
2605 		if (new_lop->nfslo_end < lop->nfslo_first) {
2606 		    /*
2607 		     * If the new lock ends before the start of the
2608 		     * current lock's range, no merge, just insert
2609 		     * the new lock.
2610 		     */
2611 		    break;
2612 		}
2613 		if (new_lop->nfslo_type == lop->nfslo_type ||
2614 		    (new_lop->nfslo_first <= lop->nfslo_first &&
2615 		     new_lop->nfslo_end >= lop->nfslo_end)) {
2616 		    /*
2617 		     * This lock can be absorbed by the new lock/unlock.
2618 		     * This happens when it covers the entire range
2619 		     * of the old lock or is contiguous
2620 		     * with the old lock and is of the same type or an
2621 		     * unlock.
2622 		     */
2623 		    if (new_lop->nfslo_type != lop->nfslo_type ||
2624 			new_lop->nfslo_first != lop->nfslo_first ||
2625 			new_lop->nfslo_end != lop->nfslo_end)
2626 			modified = 1;
2627 		    if (lop->nfslo_first < new_lop->nfslo_first)
2628 			new_lop->nfslo_first = lop->nfslo_first;
2629 		    if (lop->nfslo_end > new_lop->nfslo_end)
2630 			new_lop->nfslo_end = lop->nfslo_end;
2631 		    tlop = lop;
2632 		    lop = LIST_NEXT(lop, nfslo_list);
2633 		    nfscl_freelock(tlop, local);
2634 		    continue;
2635 		}
2636 
2637 		/*
2638 		 * All these cases are for contiguous locks that are not the
2639 		 * same type, so they can't be merged.
2640 		 */
2641 		if (new_lop->nfslo_first <= lop->nfslo_first) {
2642 		    /*
2643 		     * This case is where the new lock overlaps with the
2644 		     * first part of the old lock. Move the start of the
2645 		     * old lock to just past the end of the new lock. The
2646 		     * new lock will be inserted in front of the old, since
2647 		     * ilop hasn't been updated. (We are done now.)
2648 		     */
2649 		    if (lop->nfslo_first != new_lop->nfslo_end) {
2650 			lop->nfslo_first = new_lop->nfslo_end;
2651 			modified = 1;
2652 		    }
2653 		    break;
2654 		}
2655 		if (new_lop->nfslo_end >= lop->nfslo_end) {
2656 		    /*
2657 		     * This case is where the new lock overlaps with the
2658 		     * end of the old lock's range. Move the old lock's
2659 		     * end to just before the new lock's first and insert
2660 		     * the new lock after the old lock.
2661 		     * Might not be done yet, since the new lock could
2662 		     * overlap further locks with higher ranges.
2663 		     */
2664 		    if (lop->nfslo_end != new_lop->nfslo_first) {
2665 			lop->nfslo_end = new_lop->nfslo_first;
2666 			modified = 1;
2667 		    }
2668 		    ilop = lop;
2669 		    lop = LIST_NEXT(lop, nfslo_list);
2670 		    continue;
2671 		}
2672 		/*
2673 		 * The final case is where the new lock's range is in the
2674 		 * middle of the current lock's and splits the current lock
2675 		 * up. Use *other_lopp to handle the second part of the
2676 		 * split old lock range. (We are done now.)
2677 		 * For unlock, we use new_lop as other_lop and tmp, since
2678 		 * other_lop and new_lop are the same for this case.
2679 		 * We noted the unlock case above, so we don't need
2680 		 * new_lop->nfslo_type any longer.
2681 		 */
2682 		tmp = new_lop->nfslo_first;
2683 		if (unlock) {
2684 		    other_lop = new_lop;
2685 		    *new_lopp = NULL;
2686 		} else {
2687 		    other_lop = *other_lopp;
2688 		    *other_lopp = NULL;
2689 		}
2690 		other_lop->nfslo_first = new_lop->nfslo_end;
2691 		other_lop->nfslo_end = lop->nfslo_end;
2692 		other_lop->nfslo_type = lop->nfslo_type;
2693 		lop->nfslo_end = tmp;
2694 		nfscl_insertlock(lp, other_lop, lop, local);
2695 		ilop = lop;
2696 		modified = 1;
2697 		break;
2698 	    }
2699 	    ilop = lop;
2700 	    lop = LIST_NEXT(lop, nfslo_list);
2701 	    if (lop == NULL)
2702 		break;
2703 	}
2704 
2705 	/*
2706 	 * Insert the new lock in the list at the appropriate place.
2707 	 */
2708 	if (!unlock) {
2709 		nfscl_insertlock(lp, new_lop, ilop, local);
2710 		*new_lopp = NULL;
2711 		modified = 1;
2712 	}
2713 	return (modified);
2714 }
2715 
2716 /*
2717  * This function must be run as a kernel thread.
2718  * It does Renew Ops and recovery, when required.
2719  */
2720 void
2721 nfscl_renewthread(struct nfsclclient *clp, NFSPROC_T *p)
2722 {
2723 	struct nfsclowner *owp, *nowp;
2724 	struct nfsclopen *op;
2725 	struct nfscllockowner *lp, *nlp;
2726 	struct nfscldeleghead dh;
2727 	struct nfscldeleg *dp, *ndp;
2728 	struct ucred *cred;
2729 	u_int32_t clidrev;
2730 	int error, cbpathdown, islept, igotlock, ret, clearok;
2731 	uint32_t recover_done_time = 0;
2732 	time_t mytime;
2733 	static time_t prevsec = 0;
2734 	struct nfscllockownerfh *lfhp, *nlfhp;
2735 	struct nfscllockownerfhhead lfh;
2736 	struct nfscllayout *lyp, *nlyp;
2737 	struct nfscldevinfo *dip, *ndip;
2738 	struct nfscllayouthead rlh;
2739 	struct nfsclrecalllayout *recallp;
2740 	struct nfsclds *dsp;
2741 	bool retok;
2742 	struct mount *mp;
2743 	vnode_t vp;
2744 
2745 	cred = newnfs_getcred();
2746 	NFSLOCKCLSTATE();
2747 	clp->nfsc_flags |= NFSCLFLAGS_HASTHREAD;
2748 	mp = clp->nfsc_nmp->nm_mountp;
2749 	NFSUNLOCKCLSTATE();
2750 	for(;;) {
2751 		newnfs_setroot(cred);
2752 		cbpathdown = 0;
2753 		if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2754 			/*
2755 			 * Only allow one full recover within 1/2 of the lease
2756 			 * duration (nfsc_renew).
2757 			 * retok is value/result.  If passed in set to true,
2758 			 * it indicates only a CreateSession operation should
2759 			 * be attempted.
2760 			 * If it is returned true, it indicates that the
2761 			 * recovery only required a CreateSession.
2762 			 */
2763 			retok = true;
2764 			if (recover_done_time < NFSD_MONOSEC) {
2765 				recover_done_time = NFSD_MONOSEC +
2766 				    clp->nfsc_renew;
2767 				retok = false;
2768 			}
2769 			NFSCL_DEBUG(1, "Doing recovery, only "
2770 			    "createsession=%d\n", retok);
2771 			nfscl_recover(clp, &retok, cred, p);
2772 		}
2773 		if (clp->nfsc_expire <= NFSD_MONOSEC &&
2774 		    (clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2775 			clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2776 			clidrev = clp->nfsc_clientidrev;
2777 			error = nfsrpc_renew(clp, NULL, cred, p);
2778 			if (error == NFSERR_CBPATHDOWN)
2779 			    cbpathdown = 1;
2780 			else if (error == NFSERR_STALECLIENTID) {
2781 			    NFSLOCKCLSTATE();
2782 			    clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
2783 			    NFSUNLOCKCLSTATE();
2784 			} else if (error == NFSERR_EXPIRED)
2785 			    (void) nfscl_hasexpired(clp, clidrev, p);
2786 		}
2787 
2788 checkdsrenew:
2789 		if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2790 			/* Do renews for any DS sessions. */
2791 			NFSLOCKMNT(clp->nfsc_nmp);
2792 			/* Skip first entry, since the MDS is handled above. */
2793 			dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2794 			if (dsp != NULL)
2795 				dsp = TAILQ_NEXT(dsp, nfsclds_list);
2796 			while (dsp != NULL) {
2797 				if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2798 				    dsp->nfsclds_sess.nfsess_defunct == 0) {
2799 					dsp->nfsclds_expire = NFSD_MONOSEC +
2800 					    clp->nfsc_renew;
2801 					NFSUNLOCKMNT(clp->nfsc_nmp);
2802 					(void)nfsrpc_renew(clp, dsp, cred, p);
2803 					goto checkdsrenew;
2804 				}
2805 				dsp = TAILQ_NEXT(dsp, nfsclds_list);
2806 			}
2807 			NFSUNLOCKMNT(clp->nfsc_nmp);
2808 		}
2809 
2810 		TAILQ_INIT(&dh);
2811 		NFSLOCKCLSTATE();
2812 		if (cbpathdown)
2813 			/* It's a Total Recall! */
2814 			nfscl_totalrecall(clp);
2815 
2816 		/*
2817 		 * Now, handle defunct owners.
2818 		 */
2819 		LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2820 			if (LIST_EMPTY(&owp->nfsow_open)) {
2821 				if (owp->nfsow_defunct != 0)
2822 					nfscl_freeopenowner(owp, 0);
2823 			}
2824 		}
2825 
2826 		/*
2827 		 * Do the recall on any delegations. To avoid trouble, always
2828 		 * come back up here after having slept.
2829 		 */
2830 		igotlock = 0;
2831 tryagain:
2832 		dp = TAILQ_FIRST(&clp->nfsc_deleg);
2833 		while (dp != NULL) {
2834 			ndp = TAILQ_NEXT(dp, nfsdl_list);
2835 			if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2836 				/*
2837 				 * Wait for outstanding I/O ops to be done.
2838 				 */
2839 				if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2840 				    if (igotlock) {
2841 					nfsv4_unlock(&clp->nfsc_lock, 0);
2842 					igotlock = 0;
2843 				    }
2844 				    dp->nfsdl_rwlock.nfslock_lock |=
2845 					NFSV4LOCK_WANTED;
2846 				    msleep(&dp->nfsdl_rwlock,
2847 					NFSCLSTATEMUTEXPTR, PVFS, "nfscld",
2848 					5 * hz);
2849 				    if (NFSCL_FORCEDISM(mp))
2850 					goto terminate;
2851 				    goto tryagain;
2852 				}
2853 				while (!igotlock) {
2854 				    igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2855 					&islept, NFSCLSTATEMUTEXPTR, mp);
2856 				    if (igotlock == 0 && NFSCL_FORCEDISM(mp))
2857 					goto terminate;
2858 				    if (islept)
2859 					goto tryagain;
2860 				}
2861 				NFSUNLOCKCLSTATE();
2862 				newnfs_copycred(&dp->nfsdl_cred, cred);
2863 				ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2864 				    NULL, cred, p, 1, &vp);
2865 				if (!ret) {
2866 				    nfscl_cleandeleg(dp);
2867 				    TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2868 					nfsdl_list);
2869 				    LIST_REMOVE(dp, nfsdl_hash);
2870 				    TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2871 				    nfscl_delegcnt--;
2872 				    nfsstatsv1.cldelegates--;
2873 				}
2874 				NFSLOCKCLSTATE();
2875 				/*
2876 				 * The nfsc_lock must be released before doing
2877 				 * vrele(), since it might call nfs_inactive().
2878 				 * For the unlikely case where the vnode failed
2879 				 * to be acquired by nfscl_recalldeleg(), a
2880 				 * VOP_RECLAIM() should be in progress and it
2881 				 * will return the delegation.
2882 				 */
2883 				nfsv4_unlock(&clp->nfsc_lock, 0);
2884 				igotlock = 0;
2885 				if (vp != NULL) {
2886 					NFSUNLOCKCLSTATE();
2887 					vrele(vp);
2888 					NFSLOCKCLSTATE();
2889 				}
2890 				goto tryagain;
2891 			}
2892 			dp = ndp;
2893 		}
2894 
2895 		/*
2896 		 * Clear out old delegations, if we are above the high water
2897 		 * mark. Only clear out ones with no state related to them.
2898 		 * The tailq list is in LRU order.
2899 		 */
2900 		dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2901 		while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2902 		    ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2903 		    if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2904 			dp->nfsdl_rwlock.nfslock_lock == 0 &&
2905 			dp->nfsdl_timestamp < NFSD_MONOSEC &&
2906 			(dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_ZAPPED |
2907 			  NFSCLDL_NEEDRECLAIM | NFSCLDL_DELEGRET)) == 0) {
2908 			clearok = 1;
2909 			LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2910 			    op = LIST_FIRST(&owp->nfsow_open);
2911 			    if (op != NULL) {
2912 				clearok = 0;
2913 				break;
2914 			    }
2915 			}
2916 			if (clearok) {
2917 			    LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2918 				if (!LIST_EMPTY(&lp->nfsl_lock)) {
2919 				    clearok = 0;
2920 				    break;
2921 				}
2922 			    }
2923 			}
2924 			if (clearok) {
2925 			    TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2926 			    LIST_REMOVE(dp, nfsdl_hash);
2927 			    TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2928 			    nfscl_delegcnt--;
2929 			    nfsstatsv1.cldelegates--;
2930 			}
2931 		    }
2932 		    dp = ndp;
2933 		}
2934 		if (igotlock)
2935 			nfsv4_unlock(&clp->nfsc_lock, 0);
2936 
2937 		/*
2938 		 * Do the recall on any layouts. To avoid trouble, always
2939 		 * come back up here after having slept.
2940 		 */
2941 		TAILQ_INIT(&rlh);
2942 tryagain2:
2943 		TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2944 			if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2945 				/*
2946 				 * Wait for outstanding I/O ops to be done.
2947 				 */
2948 				if (lyp->nfsly_lock.nfslock_usecnt > 0 ||
2949 				    (lyp->nfsly_lock.nfslock_lock &
2950 				     NFSV4LOCK_LOCK) != 0) {
2951 					lyp->nfsly_lock.nfslock_lock |=
2952 					    NFSV4LOCK_WANTED;
2953 					msleep(&lyp->nfsly_lock.nfslock_lock,
2954 					    NFSCLSTATEMUTEXPTR, PVFS, "nfslyp",
2955 					    5 * hz);
2956 					if (NFSCL_FORCEDISM(mp))
2957 					    goto terminate;
2958 					goto tryagain2;
2959 				}
2960 				/* Move the layout to the recall list. */
2961 				TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2962 				    nfsly_list);
2963 				LIST_REMOVE(lyp, nfsly_hash);
2964 				TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2965 
2966 				/* Handle any layout commits. */
2967 				if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2968 				    (lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2969 					lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2970 					NFSUNLOCKCLSTATE();
2971 					NFSCL_DEBUG(3, "do layoutcommit\n");
2972 					nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2973 					    cred, p);
2974 					NFSLOCKCLSTATE();
2975 					goto tryagain2;
2976 				}
2977 			}
2978 		}
2979 
2980 		/* Now, look for stale layouts. */
2981 		lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2982 		while (lyp != NULL) {
2983 			nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2984 			if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2985 			    (lyp->nfsly_flags & (NFSLY_RECALL |
2986 			     NFSLY_RETONCLOSE)) == 0 &&
2987 			    lyp->nfsly_lock.nfslock_usecnt == 0 &&
2988 			    lyp->nfsly_lock.nfslock_lock == 0) {
2989 				NFSCL_DEBUG(4, "ret stale lay=%d\n",
2990 				    nfscl_layoutcnt);
2991 				recallp = malloc(sizeof(*recallp),
2992 				    M_NFSLAYRECALL, M_NOWAIT);
2993 				if (recallp == NULL)
2994 					break;
2995 				(void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2996 				    lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2997 				    lyp->nfsly_stateid.seqid, 0, 0, NULL,
2998 				    recallp);
2999 			}
3000 			lyp = nlyp;
3001 		}
3002 
3003 		/*
3004 		 * Free up any unreferenced device info structures.
3005 		 */
3006 		LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
3007 			if (dip->nfsdi_layoutrefs == 0 &&
3008 			    dip->nfsdi_refcnt == 0) {
3009 				NFSCL_DEBUG(4, "freeing devinfo\n");
3010 				LIST_REMOVE(dip, nfsdi_list);
3011 				nfscl_freedevinfo(dip);
3012 			}
3013 		}
3014 		NFSUNLOCKCLSTATE();
3015 
3016 		/* Do layout return(s), as required. */
3017 		TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
3018 			TAILQ_REMOVE(&rlh, lyp, nfsly_list);
3019 			NFSCL_DEBUG(4, "ret layout\n");
3020 			nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
3021 			if ((lyp->nfsly_flags & NFSLY_RETONCLOSE) != 0) {
3022 				NFSLOCKCLSTATE();
3023 				lyp->nfsly_flags |= NFSLY_RETURNED;
3024 				wakeup(lyp);
3025 				NFSUNLOCKCLSTATE();
3026 			} else
3027 				nfscl_freelayout(lyp);
3028 		}
3029 
3030 		/*
3031 		 * Delegreturn any delegations cleaned out or recalled.
3032 		 */
3033 		TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
3034 			newnfs_copycred(&dp->nfsdl_cred, cred);
3035 			(void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3036 			TAILQ_REMOVE(&dh, dp, nfsdl_list);
3037 			free(dp, M_NFSCLDELEG);
3038 		}
3039 
3040 		SLIST_INIT(&lfh);
3041 		/*
3042 		 * Call nfscl_cleanupkext() once per second to check for
3043 		 * open/lock owners where the process has exited.
3044 		 */
3045 		mytime = NFSD_MONOSEC;
3046 		if (prevsec != mytime) {
3047 			prevsec = mytime;
3048 			nfscl_cleanupkext(clp, &lfh);
3049 		}
3050 
3051 		/*
3052 		 * Do a ReleaseLockOwner for all lock owners where the
3053 		 * associated process no longer exists, as found by
3054 		 * nfscl_cleanupkext().
3055 		 */
3056 		newnfs_setroot(cred);
3057 		SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
3058 			LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
3059 			    nlp) {
3060 				(void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
3061 				    lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
3062 				    p);
3063 				nfscl_freelockowner(lp, 0);
3064 			}
3065 			free(lfhp, M_TEMP);
3066 		}
3067 		SLIST_INIT(&lfh);
3068 
3069 		NFSLOCKCLSTATE();
3070 		if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
3071 			(void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
3072 			    hz);
3073 terminate:
3074 		if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
3075 			clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
3076 			NFSUNLOCKCLSTATE();
3077 			NFSFREECRED(cred);
3078 			wakeup((caddr_t)clp);
3079 			return;
3080 		}
3081 		NFSUNLOCKCLSTATE();
3082 	}
3083 }
3084 
3085 /*
3086  * Initiate state recovery. Called when NFSERR_STALECLIENTID,
3087  * NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
3088  */
3089 void
3090 nfscl_initiate_recovery(struct nfsclclient *clp)
3091 {
3092 
3093 	if (clp == NULL)
3094 		return;
3095 	NFSLOCKCLSTATE();
3096 	clp->nfsc_flags |= NFSCLFLAGS_RECOVER;
3097 	NFSUNLOCKCLSTATE();
3098 	wakeup((caddr_t)clp);
3099 }
3100 
3101 /*
3102  * Dump out the state stuff for debugging.
3103  */
3104 void
3105 nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
3106     int lockowner, int locks)
3107 {
3108 	struct nfsclclient *clp;
3109 	struct nfsclowner *owp;
3110 	struct nfsclopen *op;
3111 	struct nfscllockowner *lp;
3112 	struct nfscllock *lop;
3113 	struct nfscldeleg *dp;
3114 
3115 	clp = nmp->nm_clp;
3116 	if (clp == NULL) {
3117 		printf("nfscl dumpstate NULL clp\n");
3118 		return;
3119 	}
3120 	NFSLOCKCLSTATE();
3121 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
3122 	  LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3123 	    if (openowner && !LIST_EMPTY(&owp->nfsow_open))
3124 		printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
3125 		    owp->nfsow_owner[0], owp->nfsow_owner[1],
3126 		    owp->nfsow_owner[2], owp->nfsow_owner[3],
3127 		    owp->nfsow_seqid);
3128 	    LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3129 		if (opens)
3130 		    printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
3131 			op->nfso_stateid.other[0], op->nfso_stateid.other[1],
3132 			op->nfso_stateid.other[2], op->nfso_opencnt,
3133 			op->nfso_fh[12]);
3134 		LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
3135 		    if (lockowner)
3136 			printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3137 			    lp->nfsl_owner[0], lp->nfsl_owner[1],
3138 			    lp->nfsl_owner[2], lp->nfsl_owner[3],
3139 			    lp->nfsl_seqid,
3140 			    lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3141 			    lp->nfsl_stateid.other[2]);
3142 		    LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3143 			if (locks)
3144 #ifdef __FreeBSD__
3145 			    printf("lck typ=%d fst=%ju end=%ju\n",
3146 				lop->nfslo_type, (intmax_t)lop->nfslo_first,
3147 				(intmax_t)lop->nfslo_end);
3148 #else
3149 			    printf("lck typ=%d fst=%qd end=%qd\n",
3150 				lop->nfslo_type, lop->nfslo_first,
3151 				lop->nfslo_end);
3152 #endif
3153 		    }
3154 		}
3155 	    }
3156 	  }
3157 	}
3158 	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3159 	    if (openowner && !LIST_EMPTY(&owp->nfsow_open))
3160 		printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
3161 		    owp->nfsow_owner[0], owp->nfsow_owner[1],
3162 		    owp->nfsow_owner[2], owp->nfsow_owner[3],
3163 		    owp->nfsow_seqid);
3164 	    LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3165 		if (opens)
3166 		    printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
3167 			op->nfso_stateid.other[0], op->nfso_stateid.other[1],
3168 			op->nfso_stateid.other[2], op->nfso_opencnt,
3169 			op->nfso_fh[12]);
3170 		LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
3171 		    if (lockowner)
3172 			printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
3173 			    lp->nfsl_owner[0], lp->nfsl_owner[1],
3174 			    lp->nfsl_owner[2], lp->nfsl_owner[3],
3175 			    lp->nfsl_seqid,
3176 			    lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
3177 			    lp->nfsl_stateid.other[2]);
3178 		    LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3179 			if (locks)
3180 #ifdef __FreeBSD__
3181 			    printf("lck typ=%d fst=%ju end=%ju\n",
3182 				lop->nfslo_type, (intmax_t)lop->nfslo_first,
3183 				(intmax_t)lop->nfslo_end);
3184 #else
3185 			    printf("lck typ=%d fst=%qd end=%qd\n",
3186 				lop->nfslo_type, lop->nfslo_first,
3187 				lop->nfslo_end);
3188 #endif
3189 		    }
3190 		}
3191 	    }
3192 	}
3193 	NFSUNLOCKCLSTATE();
3194 }
3195 
3196 /*
3197  * Check for duplicate open owners and opens.
3198  * (Only used as a diagnostic aid.)
3199  */
3200 void
3201 nfscl_dupopen(vnode_t vp, int dupopens)
3202 {
3203 	struct nfsclclient *clp;
3204 	struct nfsclowner *owp, *owp2;
3205 	struct nfsclopen *op, *op2;
3206 	struct nfsfh *nfhp;
3207 
3208 	clp = VFSTONFS(vp->v_mount)->nm_clp;
3209 	if (clp == NULL) {
3210 		printf("nfscl dupopen NULL clp\n");
3211 		return;
3212 	}
3213 	nfhp = VTONFS(vp)->n_fhp;
3214 	NFSLOCKCLSTATE();
3215 
3216 	/*
3217 	 * First, search for duplicate owners.
3218 	 * These should never happen!
3219 	 */
3220 	LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3221 	    LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3222 		if (owp != owp2 &&
3223 		    !NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
3224 		    NFSV4CL_LOCKNAMELEN)) {
3225 			NFSUNLOCKCLSTATE();
3226 			printf("DUP OWNER\n");
3227 			nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3228 			return;
3229 		}
3230 	    }
3231 	}
3232 
3233 	/*
3234 	 * Now, search for duplicate stateids.
3235 	 * These shouldn't happen, either.
3236 	 */
3237 	LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3238 	    LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3239 		LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3240 		    LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3241 			if (op != op2 &&
3242 			    (op->nfso_stateid.other[0] != 0 ||
3243 			     op->nfso_stateid.other[1] != 0 ||
3244 			     op->nfso_stateid.other[2] != 0) &&
3245 			    op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3246 			    op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3247 			    op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3248 			    NFSUNLOCKCLSTATE();
3249 			    printf("DUP STATEID\n");
3250 			    nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0, 0);
3251 			    return;
3252 			}
3253 		    }
3254 		}
3255 	    }
3256 	}
3257 
3258 	/*
3259 	 * Now search for duplicate opens.
3260 	 * Duplicate opens for the same owner
3261 	 * should never occur. Other duplicates are
3262 	 * possible and are checked for if "dupopens"
3263 	 * is true.
3264 	 */
3265 	LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3266 	    LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3267 		if (nfhp->nfh_len == op2->nfso_fhlen &&
3268 		    !NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3269 		    LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3270 			LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3271 			    if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3272 				!NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3273 				(!NFSBCMP(op->nfso_own->nfsow_owner,
3274 				 op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) ||
3275 				 dupopens)) {
3276 				if (!NFSBCMP(op->nfso_own->nfsow_owner,
3277 				    op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3278 				    NFSUNLOCKCLSTATE();
3279 				    printf("BADDUP OPEN\n");
3280 				} else {
3281 				    NFSUNLOCKCLSTATE();
3282 				    printf("DUP OPEN\n");
3283 				}
3284 				nfscl_dumpstate(VFSTONFS(vp->v_mount), 1, 1, 0,
3285 				    0);
3286 				return;
3287 			    }
3288 			}
3289 		    }
3290 		}
3291 	    }
3292 	}
3293 	NFSUNLOCKCLSTATE();
3294 }
3295 
3296 /*
3297  * During close, find an open that needs to be dereferenced and
3298  * dereference it. If there are no more opens for this file,
3299  * log a message to that effect.
3300  * Opens aren't actually Close'd until VOP_INACTIVE() is performed
3301  * on the file's vnode.
3302  * This is the safe way, since it is difficult to identify
3303  * which open the close is for and I/O can be performed after the
3304  * close(2) system call when a file is mmap'd.
3305  * If it returns 0 for success, there will be a referenced
3306  * clp returned via clpp.
3307  */
3308 int
3309 nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3310 {
3311 	struct nfsclclient *clp;
3312 	struct nfsclowner *owp;
3313 	struct nfsclopen *op;
3314 	struct nfscldeleg *dp;
3315 	struct nfsfh *nfhp;
3316 	int error, notdecr;
3317 
3318 	error = nfscl_getcl(vp->v_mount, NULL, NULL, false, true, &clp);
3319 	if (error)
3320 		return (error);
3321 	*clpp = clp;
3322 
3323 	nfhp = VTONFS(vp)->n_fhp;
3324 	notdecr = 1;
3325 	NFSLOCKCLSTATE();
3326 	/*
3327 	 * First, look for one under a delegation that was locally issued
3328 	 * and just decrement the opencnt for it. Since all my Opens against
3329 	 * the server are DENY_NONE, I don't see a problem with hanging
3330 	 * onto them. (It is much easier to use one of the extant Opens
3331 	 * that I already have on the server when a Delegation is recalled
3332 	 * than to do fresh Opens.) Someday, I might need to rethink this, but.
3333 	 */
3334 	dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3335 	if (dp != NULL) {
3336 		LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3337 			op = LIST_FIRST(&owp->nfsow_open);
3338 			if (op != NULL) {
3339 				/*
3340 				 * Since a delegation is for a file, there
3341 				 * should never be more than one open for
3342 				 * each openowner.
3343 				 */
3344 				if (LIST_NEXT(op, nfso_list) != NULL)
3345 					panic("nfscdeleg opens");
3346 				if (notdecr && op->nfso_opencnt > 0) {
3347 					notdecr = 0;
3348 					op->nfso_opencnt--;
3349 					break;
3350 				}
3351 			}
3352 		}
3353 	}
3354 
3355 	/* Now process the opens against the server. */
3356 	LIST_FOREACH(op, NFSCLOPENHASH(clp, nfhp->nfh_fh, nfhp->nfh_len),
3357 	    nfso_hash) {
3358 		if (op->nfso_fhlen == nfhp->nfh_len &&
3359 		    !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3360 		    nfhp->nfh_len)) {
3361 			/* Found an open, decrement cnt if possible */
3362 			if (notdecr && op->nfso_opencnt > 0) {
3363 				notdecr = 0;
3364 				op->nfso_opencnt--;
3365 			}
3366 			/*
3367 			 * There are more opens, so just return.
3368 			 */
3369 			if (op->nfso_opencnt > 0) {
3370 				NFSUNLOCKCLSTATE();
3371 				return (0);
3372 			}
3373 		}
3374 	}
3375 	NFSUNLOCKCLSTATE();
3376 	if (notdecr)
3377 		printf("nfscl: never fnd open\n");
3378 	return (0);
3379 }
3380 
3381 int
3382 nfscl_doclose(vnode_t vp, struct nfsclclient **clpp, NFSPROC_T *p)
3383 {
3384 	struct nfsclclient *clp;
3385 	struct nfsmount *nmp;
3386 	struct nfsclowner *owp, *nowp;
3387 	struct nfsclopen *op, *nop;
3388 	struct nfsclopenhead delayed;
3389 	struct nfscldeleg *dp;
3390 	struct nfsfh *nfhp;
3391 	struct nfsclrecalllayout *recallp;
3392 	struct nfscllayout *lyp;
3393 	int error;
3394 
3395 	error = nfscl_getcl(vp->v_mount, NULL, NULL, false, true, &clp);
3396 	if (error)
3397 		return (error);
3398 	*clpp = clp;
3399 
3400 	nmp = VFSTONFS(vp->v_mount);
3401 	nfhp = VTONFS(vp)->n_fhp;
3402 	recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3403 	NFSLOCKCLSTATE();
3404 	/*
3405 	 * First get rid of the local Open structures, which should be no
3406 	 * longer in use.
3407 	 */
3408 	dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3409 	if (dp != NULL) {
3410 		LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3411 			op = LIST_FIRST(&owp->nfsow_open);
3412 			if (op != NULL) {
3413 				KASSERT((op->nfso_opencnt == 0),
3414 				    ("nfscl: bad open cnt on deleg"));
3415 				nfscl_freeopen(op, 1, true);
3416 			}
3417 			nfscl_freeopenowner(owp, 1);
3418 		}
3419 	}
3420 
3421 	/* Return any layouts marked return on close. */
3422 	nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp,
3423 	    &lyp);
3424 
3425 	/* Now process the opens against the server. */
3426 	LIST_INIT(&delayed);
3427 lookformore:
3428 	LIST_FOREACH(op, NFSCLOPENHASH(clp, nfhp->nfh_fh, nfhp->nfh_len),
3429 	    nfso_hash) {
3430 		if (op->nfso_fhlen == nfhp->nfh_len &&
3431 		    !NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3432 		    nfhp->nfh_len)) {
3433 			/* Found an open, close it. */
3434 #ifdef DIAGNOSTIC
3435 			KASSERT((op->nfso_opencnt == 0),
3436 			    ("nfscl: bad open cnt on server (%d)",
3437 			     op->nfso_opencnt));
3438 #endif
3439 			NFSUNLOCKCLSTATE();
3440 			if (NFSHASNFSV4N(nmp))
3441 				error = nfsrpc_doclose(nmp, op, p, false, true);
3442 			else
3443 				error = nfsrpc_doclose(nmp, op, p, true, true);
3444 			NFSLOCKCLSTATE();
3445 			if (error == NFSERR_DELAY) {
3446 				nfscl_unlinkopen(op);
3447 				op->nfso_own = NULL;
3448 				LIST_INSERT_HEAD(&delayed, op, nfso_list);
3449 			}
3450 			goto lookformore;
3451 		}
3452 	}
3453 	nfscl_clrelease(clp);
3454 
3455 	/* Now, wait for any layout that is returned upon close. */
3456 	if (lyp != NULL) {
3457 		while ((lyp->nfsly_flags & NFSLY_RETURNED) == 0) {
3458 			if (NFSCL_FORCEDISM(nmp->nm_mountp)) {
3459 				lyp = NULL;
3460 				break;
3461 			}
3462 			msleep(lyp, NFSCLSTATEMUTEXPTR, PZERO, "nfslroc", hz);
3463 		}
3464 		if (lyp != NULL)
3465 			nfscl_freelayout(lyp);
3466 	}
3467 
3468 	NFSUNLOCKCLSTATE();
3469 	/*
3470 	 * recallp has been set NULL by nfscl_retoncloselayout() if it was
3471 	 * used by the function, but calling free() with a NULL pointer is ok.
3472 	 */
3473 	free(recallp, M_NFSLAYRECALL);
3474 
3475 	/* Now, loop retrying the delayed closes. */
3476 	LIST_FOREACH_SAFE(op, &delayed, nfso_list, nop) {
3477 		nfsrpc_doclose(nmp, op, p, true, false);
3478 		LIST_REMOVE(op, nfso_list);
3479 		nfscl_freeopen(op, 0, false);
3480 	}
3481 	return (0);
3482 }
3483 
3484 /*
3485  * Return all delegations on this client.
3486  * (Must be called with client sleep lock.)
3487  */
3488 static void
3489 nfscl_delegreturnall(struct nfsclclient *clp, NFSPROC_T *p,
3490     struct nfscldeleghead *dhp)
3491 {
3492 	struct nfscldeleg *dp, *ndp;
3493 	struct ucred *cred;
3494 
3495 	cred = newnfs_getcred();
3496 	TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3497 		nfscl_cleandeleg(dp);
3498 		(void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3499 		if (dhp != NULL) {
3500 			nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
3501 			TAILQ_INSERT_HEAD(dhp, dp, nfsdl_list);
3502 		} else
3503 			nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
3504 	}
3505 	NFSFREECRED(cred);
3506 }
3507 
3508 /*
3509  * Return any delegation for this vp.
3510  */
3511 void
3512 nfscl_delegreturnvp(vnode_t vp, NFSPROC_T *p)
3513 {
3514 	struct nfsclclient *clp;
3515 	struct nfscldeleg *dp;
3516 	struct ucred *cred;
3517 	struct nfsnode *np;
3518 	struct nfsmount *nmp;
3519 
3520 	nmp = VFSTONFS(vp->v_mount);
3521 	NFSLOCKMNT(nmp);
3522 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
3523 		NFSUNLOCKMNT(nmp);
3524 		return;
3525 	}
3526 	NFSUNLOCKMNT(nmp);
3527 	np = VTONFS(vp);
3528 	cred = newnfs_getcred();
3529 	dp = NULL;
3530 	NFSLOCKCLSTATE();
3531 	clp = nmp->nm_clp;
3532 	if (clp != NULL)
3533 		dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
3534 		    np->n_fhp->nfh_len);
3535 	if (dp != NULL) {
3536 		nfscl_cleandeleg(dp);
3537 		nfscl_freedeleg(&clp->nfsc_deleg, dp, false);
3538 		NFSUNLOCKCLSTATE();
3539 		newnfs_copycred(&dp->nfsdl_cred, cred);
3540 		nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3541 		free(dp, M_NFSCLDELEG);
3542 	} else
3543 		NFSUNLOCKCLSTATE();
3544 	NFSFREECRED(cred);
3545 }
3546 
3547 /*
3548  * Do a callback RPC.
3549  */
3550 void
3551 nfscl_docb(struct nfsrv_descript *nd, NFSPROC_T *p)
3552 {
3553 	int clist, gotseq_ok, i, j, k, op, rcalls;
3554 	u_int32_t *tl;
3555 	struct nfsclclient *clp;
3556 	struct nfscldeleg *dp = NULL;
3557 	int numops, taglen = -1, error = 0, trunc __unused;
3558 	u_int32_t minorvers = 0, retops = 0, *retopsp = NULL, *repp, cbident;
3559 	u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3560 	vnode_t vp = NULL;
3561 	struct nfsnode *np;
3562 	struct vattr va;
3563 	struct nfsfh *nfhp;
3564 	mount_t mp;
3565 	nfsattrbit_t attrbits, rattrbits;
3566 	nfsv4stateid_t stateid;
3567 	uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3568 	uint8_t sessionid[NFSX_V4SESSIONID];
3569 	struct mbuf *rep;
3570 	struct nfscllayout *lyp;
3571 	uint64_t filesid[2], len, off;
3572 	int changed, gotone, laytype, recalltype;
3573 	uint32_t iomode;
3574 	struct nfsclrecalllayout *recallp = NULL;
3575 	struct nfsclsession *tsep;
3576 
3577 	gotseq_ok = 0;
3578 	nfsrvd_rephead(nd);
3579 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3580 	taglen = fxdr_unsigned(int, *tl);
3581 	if (taglen < 0 || taglen > NFSV4_OPAQUELIMIT) {
3582 		error = EBADRPC;
3583 		taglen = -1;
3584 		goto nfsmout;
3585 	}
3586 	if (taglen <= NFSV4_SMALLSTR)
3587 		tagstr = tag;
3588 	else
3589 		tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3590 	error = nfsrv_mtostr(nd, tagstr, taglen);
3591 	if (error) {
3592 		if (taglen > NFSV4_SMALLSTR)
3593 			free(tagstr, M_TEMP);
3594 		taglen = -1;
3595 		goto nfsmout;
3596 	}
3597 	(void) nfsm_strtom(nd, tag, taglen);
3598 	if (taglen > NFSV4_SMALLSTR) {
3599 		free(tagstr, M_TEMP);
3600 	}
3601 	NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3602 	NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3603 	minorvers = fxdr_unsigned(u_int32_t, *tl++);
3604 	if (minorvers != NFSV4_MINORVERSION &&
3605 	    minorvers != NFSV41_MINORVERSION &&
3606 	    minorvers != NFSV42_MINORVERSION)
3607 		nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3608 	cbident = fxdr_unsigned(u_int32_t, *tl++);
3609 	if (nd->nd_repstat)
3610 		numops = 0;
3611 	else
3612 		numops = fxdr_unsigned(int, *tl);
3613 	/*
3614 	 * Loop around doing the sub ops.
3615 	 */
3616 	for (i = 0; i < numops; i++) {
3617 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3618 		NFSM_BUILD(repp, u_int32_t *, 2 * NFSX_UNSIGNED);
3619 		*repp++ = *tl;
3620 		op = fxdr_unsigned(int, *tl);
3621 		nd->nd_procnum = op;
3622 		if (i == 0 && op != NFSV4OP_CBSEQUENCE && minorvers !=
3623 		    NFSV4_MINORVERSION) {
3624 		    nd->nd_repstat = NFSERR_OPNOTINSESS;
3625 		    *repp = nfscl_errmap(nd, minorvers);
3626 		    retops++;
3627 		    break;
3628 		}
3629 		if (op < NFSV4OP_CBGETATTR ||
3630 		   (op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) ||
3631 		   (op > NFSV4OP_CBNOTIFYDEVID &&
3632 		    minorvers == NFSV41_MINORVERSION) ||
3633 		   (op > NFSV4OP_CBOFFLOAD &&
3634 		    minorvers == NFSV42_MINORVERSION)) {
3635 		    nd->nd_repstat = NFSERR_OPILLEGAL;
3636 		    *repp = nfscl_errmap(nd, minorvers);
3637 		    retops++;
3638 		    break;
3639 		}
3640 		if (op < NFSV42_CBNOPS)
3641 			nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3642 		switch (op) {
3643 		case NFSV4OP_CBGETATTR:
3644 			NFSCL_DEBUG(4, "cbgetattr\n");
3645 			mp = NULL;
3646 			vp = NULL;
3647 			error = nfsm_getfh(nd, &nfhp);
3648 			if (!error)
3649 				error = nfsrv_getattrbits(nd, &attrbits,
3650 				    NULL, NULL);
3651 			if (!error) {
3652 				mp = nfscl_getmnt(minorvers, sessionid, cbident,
3653 				    &clp);
3654 				if (mp == NULL)
3655 					error = NFSERR_SERVERFAULT;
3656 			}
3657 			if (!error) {
3658 				error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3659 				    nfhp->nfh_len, p, &np);
3660 				if (!error)
3661 					vp = NFSTOV(np);
3662 			}
3663 			if (!error) {
3664 				NFSZERO_ATTRBIT(&rattrbits);
3665 				NFSLOCKCLSTATE();
3666 				dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3667 				    nfhp->nfh_len);
3668 				if (dp != NULL) {
3669 					if (NFSISSET_ATTRBIT(&attrbits,
3670 					    NFSATTRBIT_SIZE)) {
3671 						if (vp != NULL)
3672 							va.va_size = np->n_size;
3673 						else
3674 							va.va_size =
3675 							    dp->nfsdl_size;
3676 						NFSSETBIT_ATTRBIT(&rattrbits,
3677 						    NFSATTRBIT_SIZE);
3678 					}
3679 					if (NFSISSET_ATTRBIT(&attrbits,
3680 					    NFSATTRBIT_CHANGE)) {
3681 						va.va_filerev =
3682 						    dp->nfsdl_change;
3683 						if (vp == NULL ||
3684 						    (np->n_flag & NDELEGMOD))
3685 							va.va_filerev++;
3686 						NFSSETBIT_ATTRBIT(&rattrbits,
3687 						    NFSATTRBIT_CHANGE);
3688 					}
3689 				} else
3690 					error = NFSERR_SERVERFAULT;
3691 				NFSUNLOCKCLSTATE();
3692 			}
3693 			if (vp != NULL)
3694 				vrele(vp);
3695 			if (mp != NULL)
3696 				vfs_unbusy(mp);
3697 			if (nfhp != NULL)
3698 				free(nfhp, M_NFSFH);
3699 			if (!error)
3700 				(void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3701 				    NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3702 				    (uint64_t)0, NULL);
3703 			break;
3704 		case NFSV4OP_CBRECALL:
3705 			NFSCL_DEBUG(4, "cbrecall\n");
3706 			NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3707 			    NFSX_UNSIGNED);
3708 			stateid.seqid = *tl++;
3709 			NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3710 			    NFSX_STATEIDOTHER);
3711 			tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3712 			trunc = fxdr_unsigned(int, *tl);
3713 			error = nfsm_getfh(nd, &nfhp);
3714 			if (!error) {
3715 				NFSLOCKCLSTATE();
3716 				if (minorvers == NFSV4_MINORVERSION)
3717 					clp = nfscl_getclnt(cbident);
3718 				else
3719 					clp = nfscl_getclntsess(sessionid);
3720 				if (clp != NULL) {
3721 					dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3722 					    nfhp->nfh_len);
3723 					if (dp != NULL && (dp->nfsdl_flags &
3724 					    NFSCLDL_DELEGRET) == 0) {
3725 						dp->nfsdl_flags |=
3726 						    NFSCLDL_RECALL;
3727 						wakeup((caddr_t)clp);
3728 					}
3729 				} else {
3730 					error = NFSERR_SERVERFAULT;
3731 				}
3732 				NFSUNLOCKCLSTATE();
3733 			}
3734 			if (nfhp != NULL)
3735 				free(nfhp, M_NFSFH);
3736 			break;
3737 		case NFSV4OP_CBLAYOUTRECALL:
3738 			NFSCL_DEBUG(4, "cblayrec\n");
3739 			nfhp = NULL;
3740 			NFSM_DISSECT(tl, uint32_t *, 4 * NFSX_UNSIGNED);
3741 			laytype = fxdr_unsigned(int, *tl++);
3742 			iomode = fxdr_unsigned(uint32_t, *tl++);
3743 			if (newnfs_true == *tl++)
3744 				changed = 1;
3745 			else
3746 				changed = 0;
3747 			recalltype = fxdr_unsigned(int, *tl);
3748 			NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3749 			    laytype, iomode, changed, recalltype);
3750 			recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3751 			    M_WAITOK);
3752 			if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3753 			    laytype != NFSLAYOUT_FLEXFILE)
3754 				error = NFSERR_NOMATCHLAYOUT;
3755 			else if (recalltype == NFSLAYOUTRETURN_FILE) {
3756 				error = nfsm_getfh(nd, &nfhp);
3757 				NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3758 				if (error != 0)
3759 					goto nfsmout;
3760 				NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_HYPER +
3761 				    NFSX_STATEID);
3762 				off = fxdr_hyper(tl); tl += 2;
3763 				len = fxdr_hyper(tl); tl += 2;
3764 				stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3765 				NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3766 				if (minorvers == NFSV4_MINORVERSION)
3767 					error = NFSERR_NOTSUPP;
3768 				NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3769 				    (uintmax_t)off, (uintmax_t)len,
3770 				    stateid.seqid, error);
3771 				if (error == 0) {
3772 					NFSLOCKCLSTATE();
3773 					clp = nfscl_getclntsess(sessionid);
3774 					NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3775 					if (clp != NULL) {
3776 						lyp = nfscl_findlayout(clp,
3777 						    nfhp->nfh_fh,
3778 						    nfhp->nfh_len);
3779 						NFSCL_DEBUG(4, "cblyp=%p\n",
3780 						    lyp);
3781 						if (lyp != NULL &&
3782 						    (lyp->nfsly_flags &
3783 						     (NFSLY_FILES |
3784 						      NFSLY_FLEXFILE)) != 0 &&
3785 						    !NFSBCMP(stateid.other,
3786 						    lyp->nfsly_stateid.other,
3787 						    NFSX_STATEIDOTHER)) {
3788 							error =
3789 							    nfscl_layoutrecall(
3790 							    recalltype,
3791 							    lyp, iomode, off,
3792 							    len, stateid.seqid,
3793 							    0, 0, NULL,
3794 							    recallp);
3795 							if (error == 0 &&
3796 							    stateid.seqid >
3797 							    lyp->nfsly_stateid.seqid)
3798 								lyp->nfsly_stateid.seqid =
3799 								    stateid.seqid;
3800 							recallp = NULL;
3801 							wakeup(clp);
3802 							NFSCL_DEBUG(4,
3803 							    "aft layrcal=%d "
3804 							    "layseqid=%d\n",
3805 							    error,
3806 							    lyp->nfsly_stateid.seqid);
3807 						} else
3808 							error =
3809 							  NFSERR_NOMATCHLAYOUT;
3810 					} else
3811 						error = NFSERR_NOMATCHLAYOUT;
3812 					NFSUNLOCKCLSTATE();
3813 				}
3814 				free(nfhp, M_NFSFH);
3815 			} else if (recalltype == NFSLAYOUTRETURN_FSID) {
3816 				NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_HYPER);
3817 				filesid[0] = fxdr_hyper(tl); tl += 2;
3818 				filesid[1] = fxdr_hyper(tl); tl += 2;
3819 				gotone = 0;
3820 				NFSLOCKCLSTATE();
3821 				clp = nfscl_getclntsess(sessionid);
3822 				if (clp != NULL) {
3823 					TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3824 					    nfsly_list) {
3825 						if (lyp->nfsly_filesid[0] ==
3826 						    filesid[0] &&
3827 						    lyp->nfsly_filesid[1] ==
3828 						    filesid[1]) {
3829 							error =
3830 							    nfscl_layoutrecall(
3831 							    recalltype,
3832 							    lyp, iomode, 0,
3833 							    UINT64_MAX,
3834 							    lyp->nfsly_stateid.seqid,
3835 							    0, 0, NULL,
3836 							    recallp);
3837 							recallp = NULL;
3838 							gotone = 1;
3839 						}
3840 					}
3841 					if (gotone != 0)
3842 						wakeup(clp);
3843 					else
3844 						error = NFSERR_NOMATCHLAYOUT;
3845 				} else
3846 					error = NFSERR_NOMATCHLAYOUT;
3847 				NFSUNLOCKCLSTATE();
3848 			} else if (recalltype == NFSLAYOUTRETURN_ALL) {
3849 				gotone = 0;
3850 				NFSLOCKCLSTATE();
3851 				clp = nfscl_getclntsess(sessionid);
3852 				if (clp != NULL) {
3853 					TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3854 					    nfsly_list) {
3855 						error = nfscl_layoutrecall(
3856 						    recalltype, lyp, iomode, 0,
3857 						    UINT64_MAX,
3858 						    lyp->nfsly_stateid.seqid,
3859 						    0, 0, NULL, recallp);
3860 						recallp = NULL;
3861 						gotone = 1;
3862 					}
3863 					if (gotone != 0)
3864 						wakeup(clp);
3865 					else
3866 						error = NFSERR_NOMATCHLAYOUT;
3867 				} else
3868 					error = NFSERR_NOMATCHLAYOUT;
3869 				NFSUNLOCKCLSTATE();
3870 			} else
3871 				error = NFSERR_NOMATCHLAYOUT;
3872 			if (recallp != NULL) {
3873 				free(recallp, M_NFSLAYRECALL);
3874 				recallp = NULL;
3875 			}
3876 			break;
3877 		case NFSV4OP_CBSEQUENCE:
3878 			if (i != 0) {
3879 			    error = NFSERR_SEQUENCEPOS;
3880 			    break;
3881 			}
3882 			NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3883 			    5 * NFSX_UNSIGNED);
3884 			bcopy(tl, sessionid, NFSX_V4SESSIONID);
3885 			tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3886 			seqid = fxdr_unsigned(uint32_t, *tl++);
3887 			slotid = fxdr_unsigned(uint32_t, *tl++);
3888 			highslot = fxdr_unsigned(uint32_t, *tl++);
3889 			cachethis = *tl++;
3890 			/* Throw away the referring call stuff. */
3891 			clist = fxdr_unsigned(int, *tl);
3892 			for (j = 0; j < clist; j++) {
3893 				NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3894 				    NFSX_UNSIGNED);
3895 				tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3896 				rcalls = fxdr_unsigned(int, *tl);
3897 				for (k = 0; k < rcalls; k++) {
3898 					NFSM_DISSECT(tl, uint32_t *,
3899 					    2 * NFSX_UNSIGNED);
3900 				}
3901 			}
3902 			NFSLOCKCLSTATE();
3903 			clp = nfscl_getclntsess(sessionid);
3904 			if (clp == NULL)
3905 				error = NFSERR_SERVERFAULT;
3906 			if (error == 0) {
3907 				tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3908 				error = nfsv4_seqsession(seqid, slotid,
3909 				    highslot, tsep->nfsess_cbslots, &rep,
3910 				    tsep->nfsess_backslots);
3911 			}
3912 			NFSUNLOCKCLSTATE();
3913 			if (error == 0 || error == NFSERR_REPLYFROMCACHE) {
3914 				gotseq_ok = 1;
3915 				if (rep != NULL) {
3916 					/*
3917 					 * Handle a reply for a retried
3918 					 * callback.  The reply will be
3919 					 * re-inserted in the session cache
3920 					 * by the nfsv4_seqsess_cacherep() call
3921 					 * after out:
3922 					 */
3923 					KASSERT(error == NFSERR_REPLYFROMCACHE,
3924 					    ("cbsequence: non-NULL rep"));
3925 					NFSCL_DEBUG(4, "Got cbretry\n");
3926 					m_freem(nd->nd_mreq);
3927 					nd->nd_mreq = rep;
3928 					rep = NULL;
3929 					goto out;
3930 				}
3931 				NFSM_BUILD(tl, uint32_t *,
3932 				    NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3933 				bcopy(sessionid, tl, NFSX_V4SESSIONID);
3934 				tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3935 				*tl++ = txdr_unsigned(seqid);
3936 				*tl++ = txdr_unsigned(slotid);
3937 				*tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3938 				*tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3939 			}
3940 			break;
3941 		default:
3942 			if (i == 0 && minorvers != NFSV4_MINORVERSION)
3943 				error = NFSERR_OPNOTINSESS;
3944 			else {
3945 				NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3946 				error = NFSERR_NOTSUPP;
3947 			}
3948 			break;
3949 		}
3950 		if (error) {
3951 			if (error == EBADRPC || error == NFSERR_BADXDR) {
3952 				nd->nd_repstat = NFSERR_BADXDR;
3953 			} else {
3954 				nd->nd_repstat = error;
3955 			}
3956 			error = 0;
3957 		}
3958 		retops++;
3959 		if (nd->nd_repstat) {
3960 			*repp = nfscl_errmap(nd, minorvers);
3961 			break;
3962 		} else
3963 			*repp = 0;	/* NFS4_OK */
3964 	}
3965 nfsmout:
3966 	if (recallp != NULL)
3967 		free(recallp, M_NFSLAYRECALL);
3968 	if (error) {
3969 		if (error == EBADRPC || error == NFSERR_BADXDR)
3970 			nd->nd_repstat = NFSERR_BADXDR;
3971 		else
3972 			printf("nfsv4 comperr1=%d\n", error);
3973 	}
3974 	if (taglen == -1) {
3975 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3976 		*tl++ = 0;
3977 		*tl = 0;
3978 	} else {
3979 		*retopsp = txdr_unsigned(retops);
3980 	}
3981 	*nd->nd_errp = nfscl_errmap(nd, minorvers);
3982 out:
3983 	if (gotseq_ok != 0) {
3984 		rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3985 		NFSLOCKCLSTATE();
3986 		clp = nfscl_getclntsess(sessionid);
3987 		if (clp != NULL) {
3988 			tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3989 			nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3990 			    NFSERR_OK, &rep);
3991 			NFSUNLOCKCLSTATE();
3992 		} else {
3993 			NFSUNLOCKCLSTATE();
3994 			m_freem(rep);
3995 		}
3996 	}
3997 }
3998 
3999 /*
4000  * Generate the next cbident value. Basically just increment a static value
4001  * and then check that it isn't already in the list, if it has wrapped around.
4002  */
4003 static u_int32_t
4004 nfscl_nextcbident(void)
4005 {
4006 	struct nfsclclient *clp;
4007 	int matched;
4008 	static u_int32_t nextcbident = 0;
4009 	static int haswrapped = 0;
4010 
4011 	nextcbident++;
4012 	if (nextcbident == 0)
4013 		haswrapped = 1;
4014 	if (haswrapped) {
4015 		/*
4016 		 * Search the clientid list for one already using this cbident.
4017 		 */
4018 		do {
4019 			matched = 0;
4020 			NFSLOCKCLSTATE();
4021 			LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
4022 				if (clp->nfsc_cbident == nextcbident) {
4023 					matched = 1;
4024 					break;
4025 				}
4026 			}
4027 			NFSUNLOCKCLSTATE();
4028 			if (matched == 1)
4029 				nextcbident++;
4030 		} while (matched);
4031 	}
4032 	return (nextcbident);
4033 }
4034 
4035 /*
4036  * Get the mount point related to a given cbident or session and busy it.
4037  */
4038 static mount_t
4039 nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
4040     struct nfsclclient **clpp)
4041 {
4042 	struct nfsclclient *clp;
4043 	mount_t mp;
4044 	int error;
4045 	struct nfsclsession *tsep;
4046 
4047 	*clpp = NULL;
4048 	NFSLOCKCLSTATE();
4049 	LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
4050 		tsep = nfsmnt_mdssession(clp->nfsc_nmp);
4051 		if (minorvers == NFSV4_MINORVERSION) {
4052 			if (clp->nfsc_cbident == cbident)
4053 				break;
4054 		} else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
4055 		    NFSX_V4SESSIONID))
4056 			break;
4057 	}
4058 	if (clp == NULL) {
4059 		NFSUNLOCKCLSTATE();
4060 		return (NULL);
4061 	}
4062 	mp = clp->nfsc_nmp->nm_mountp;
4063 	vfs_ref(mp);
4064 	NFSUNLOCKCLSTATE();
4065 	error = vfs_busy(mp, 0);
4066 	vfs_rel(mp);
4067 	if (error != 0)
4068 		return (NULL);
4069 	*clpp = clp;
4070 	return (mp);
4071 }
4072 
4073 /*
4074  * Get the clientid pointer related to a given cbident.
4075  */
4076 static struct nfsclclient *
4077 nfscl_getclnt(u_int32_t cbident)
4078 {
4079 	struct nfsclclient *clp;
4080 
4081 	LIST_FOREACH(clp, &nfsclhead, nfsc_list)
4082 		if (clp->nfsc_cbident == cbident)
4083 			break;
4084 	return (clp);
4085 }
4086 
4087 /*
4088  * Get the clientid pointer related to a given sessionid.
4089  */
4090 static struct nfsclclient *
4091 nfscl_getclntsess(uint8_t *sessionid)
4092 {
4093 	struct nfsclclient *clp;
4094 	struct nfsclsession *tsep;
4095 
4096 	LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
4097 		tsep = nfsmnt_mdssession(clp->nfsc_nmp);
4098 		if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
4099 		    NFSX_V4SESSIONID))
4100 			break;
4101 	}
4102 	return (clp);
4103 }
4104 
4105 /*
4106  * Search for a lock conflict locally on the client. A conflict occurs if
4107  * - not same owner and overlapping byte range and at least one of them is
4108  *   a write lock or this is an unlock.
4109  */
4110 static int
4111 nfscl_localconflict(struct nfsclclient *clp, u_int8_t *fhp, int fhlen,
4112     struct nfscllock *nlop, u_int8_t *own, struct nfscldeleg *dp,
4113     struct nfscllock **lopp)
4114 {
4115 	struct nfsclopen *op;
4116 	int ret;
4117 
4118 	if (dp != NULL) {
4119 		ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
4120 		if (ret)
4121 			return (ret);
4122 	}
4123 	LIST_FOREACH(op, NFSCLOPENHASH(clp, fhp, fhlen), nfso_hash) {
4124 		if (op->nfso_fhlen == fhlen &&
4125 		    !NFSBCMP(op->nfso_fh, fhp, fhlen)) {
4126 			ret = nfscl_checkconflict(&op->nfso_lock, nlop,
4127 			    own, lopp);
4128 			if (ret)
4129 				return (ret);
4130 		}
4131 	}
4132 	return (0);
4133 }
4134 
4135 static int
4136 nfscl_checkconflict(struct nfscllockownerhead *lhp, struct nfscllock *nlop,
4137     u_int8_t *own, struct nfscllock **lopp)
4138 {
4139 	struct nfscllockowner *lp;
4140 	struct nfscllock *lop;
4141 
4142 	LIST_FOREACH(lp, lhp, nfsl_list) {
4143 		if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
4144 			LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
4145 				if (lop->nfslo_first >= nlop->nfslo_end)
4146 					break;
4147 				if (lop->nfslo_end <= nlop->nfslo_first)
4148 					continue;
4149 				if (lop->nfslo_type == F_WRLCK ||
4150 				    nlop->nfslo_type == F_WRLCK ||
4151 				    nlop->nfslo_type == F_UNLCK) {
4152 					if (lopp != NULL)
4153 						*lopp = lop;
4154 					return (NFSERR_DENIED);
4155 				}
4156 			}
4157 		}
4158 	}
4159 	return (0);
4160 }
4161 
4162 /*
4163  * Check for a local conflicting lock.
4164  */
4165 int
4166 nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
4167     u_int64_t len, struct flock *fl, NFSPROC_T *p, void *id, int flags)
4168 {
4169 	struct nfscllock *lop, nlck;
4170 	struct nfscldeleg *dp;
4171 	struct nfsnode *np;
4172 	u_int8_t own[NFSV4CL_LOCKNAMELEN];
4173 	int error;
4174 
4175 	nlck.nfslo_type = fl->l_type;
4176 	nlck.nfslo_first = off;
4177 	if (len == NFS64BITSSET) {
4178 		nlck.nfslo_end = NFS64BITSSET;
4179 	} else {
4180 		nlck.nfslo_end = off + len;
4181 		if (nlck.nfslo_end <= nlck.nfslo_first)
4182 			return (NFSERR_INVAL);
4183 	}
4184 	np = VTONFS(vp);
4185 	nfscl_filllockowner(id, own, flags);
4186 	NFSLOCKCLSTATE();
4187 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4188 	error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
4189 	    &nlck, own, dp, &lop);
4190 	if (error != 0) {
4191 		fl->l_whence = SEEK_SET;
4192 		fl->l_start = lop->nfslo_first;
4193 		if (lop->nfslo_end == NFS64BITSSET)
4194 			fl->l_len = 0;
4195 		else
4196 			fl->l_len = lop->nfslo_end - lop->nfslo_first;
4197 		fl->l_pid = (pid_t)0;
4198 		fl->l_type = lop->nfslo_type;
4199 		error = -1;			/* no RPC required */
4200 	} else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) ||
4201 	    fl->l_type == F_RDLCK)) {
4202 		/*
4203 		 * The delegation ensures that there isn't a conflicting
4204 		 * lock on the server, so return -1 to indicate an RPC
4205 		 * isn't required.
4206 		 */
4207 		fl->l_type = F_UNLCK;
4208 		error = -1;
4209 	}
4210 	NFSUNLOCKCLSTATE();
4211 	return (error);
4212 }
4213 
4214 /*
4215  * Handle Recall of a delegation.
4216  * The clp must be exclusive locked when this is called.
4217  */
4218 static int
4219 nfscl_recalldeleg(struct nfsclclient *clp, struct nfsmount *nmp,
4220     struct nfscldeleg *dp, vnode_t vp, struct ucred *cred, NFSPROC_T *p,
4221     int called_from_renewthread, vnode_t *vpp)
4222 {
4223 	struct nfsclowner *owp, *lowp, *nowp;
4224 	struct nfsclopen *op, *lop;
4225 	struct nfscllockowner *lp;
4226 	struct nfscllock *lckp;
4227 	struct nfsnode *np;
4228 	int error = 0, ret;
4229 
4230 	if (vp == NULL) {
4231 		KASSERT(vpp != NULL, ("nfscl_recalldeleg: vpp NULL"));
4232 		*vpp = NULL;
4233 		/*
4234 		 * First, get a vnode for the file. This is needed to do RPCs.
4235 		 */
4236 		ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
4237 		    dp->nfsdl_fhlen, p, &np);
4238 		if (ret) {
4239 			/*
4240 			 * File isn't open, so nothing to move over to the
4241 			 * server.
4242 			 */
4243 			return (0);
4244 		}
4245 		vp = NFSTOV(np);
4246 		*vpp = vp;
4247 	} else {
4248 		np = VTONFS(vp);
4249 	}
4250 	dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
4251 
4252 	/*
4253 	 * Ok, if it's a write delegation, flush data to the server, so
4254 	 * that close/open consistency is retained.
4255 	 */
4256 	ret = 0;
4257 	NFSLOCKNODE(np);
4258 	if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
4259 		np->n_flag |= NDELEGRECALL;
4260 		NFSUNLOCKNODE(np);
4261 		ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
4262 		NFSLOCKNODE(np);
4263 		np->n_flag &= ~NDELEGRECALL;
4264 	}
4265 	NFSINVALATTRCACHE(np);
4266 	NFSUNLOCKNODE(np);
4267 	if (ret == EIO && called_from_renewthread != 0) {
4268 		/*
4269 		 * If the flush failed with EIO for the renew thread,
4270 		 * return now, so that the dirty buffer will be flushed
4271 		 * later.
4272 		 */
4273 		return (ret);
4274 	}
4275 
4276 	/*
4277 	 * Now, for each openowner with opens issued locally, move them
4278 	 * over to state against the server.
4279 	 */
4280 	LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
4281 		lop = LIST_FIRST(&lowp->nfsow_open);
4282 		if (lop != NULL) {
4283 			if (LIST_NEXT(lop, nfso_list) != NULL)
4284 				panic("nfsdlg mult opens");
4285 			/*
4286 			 * Look for the same openowner against the server.
4287 			 */
4288 			LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
4289 				if (!NFSBCMP(lowp->nfsow_owner,
4290 				    owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
4291 					newnfs_copycred(&dp->nfsdl_cred, cred);
4292 					ret = nfscl_moveopen(vp, clp, nmp, lop,
4293 					    owp, dp, cred, p);
4294 					if (ret == NFSERR_STALECLIENTID ||
4295 					    ret == NFSERR_STALEDONTRECOVER ||
4296 					    ret == NFSERR_BADSESSION)
4297 						return (ret);
4298 					if (ret) {
4299 						nfscl_freeopen(lop, 1, true);
4300 						if (!error)
4301 							error = ret;
4302 					}
4303 					break;
4304 				}
4305 			}
4306 
4307 			/*
4308 			 * If no openowner found, create one and get an open
4309 			 * for it.
4310 			 */
4311 			if (owp == NULL) {
4312 				nowp = malloc(
4313 				    sizeof (struct nfsclowner), M_NFSCLOWNER,
4314 				    M_WAITOK);
4315 				nfscl_newopen(clp, NULL, &owp, &nowp, &op,
4316 				    NULL, lowp->nfsow_owner, dp->nfsdl_fh,
4317 				    dp->nfsdl_fhlen, NULL, NULL);
4318 				newnfs_copycred(&dp->nfsdl_cred, cred);
4319 				ret = nfscl_moveopen(vp, clp, nmp, lop,
4320 				    owp, dp, cred, p);
4321 				if (ret) {
4322 					nfscl_freeopenowner(owp, 0);
4323 					if (ret == NFSERR_STALECLIENTID ||
4324 					    ret == NFSERR_STALEDONTRECOVER ||
4325 					    ret == NFSERR_BADSESSION)
4326 						return (ret);
4327 					if (ret) {
4328 						nfscl_freeopen(lop, 1, true);
4329 						if (!error)
4330 							error = ret;
4331 					}
4332 				}
4333 			}
4334 		}
4335 	}
4336 
4337 	/*
4338 	 * Now, get byte range locks for any locks done locally.
4339 	 */
4340 	LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4341 		LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4342 			newnfs_copycred(&dp->nfsdl_cred, cred);
4343 			ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4344 			if (ret == NFSERR_STALESTATEID ||
4345 			    ret == NFSERR_STALEDONTRECOVER ||
4346 			    ret == NFSERR_STALECLIENTID ||
4347 			    ret == NFSERR_BADSESSION)
4348 				return (ret);
4349 			if (ret && !error)
4350 				error = ret;
4351 		}
4352 	}
4353 	return (error);
4354 }
4355 
4356 /*
4357  * Move a locally issued open over to an owner on the state list.
4358  * SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4359  * returns with it unlocked.
4360  */
4361 static int
4362 nfscl_moveopen(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4363     struct nfsclopen *lop, struct nfsclowner *owp, struct nfscldeleg *dp,
4364     struct ucred *cred, NFSPROC_T *p)
4365 {
4366 	struct nfsclopen *op, *nop;
4367 	struct nfscldeleg *ndp;
4368 	struct nfsnode *np;
4369 	int error = 0, newone;
4370 
4371 	/*
4372 	 * First, look for an appropriate open, If found, just increment the
4373 	 * opencnt in it.
4374 	 */
4375 	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4376 		if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4377 		    op->nfso_fhlen == lop->nfso_fhlen &&
4378 		    !NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4379 			op->nfso_opencnt += lop->nfso_opencnt;
4380 			nfscl_freeopen(lop, 1, true);
4381 			return (0);
4382 		}
4383 	}
4384 
4385 	/* No appropriate open, so we have to do one against the server. */
4386 	np = VTONFS(vp);
4387 	nop = malloc(sizeof (struct nfsclopen) +
4388 	    lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4389 	nop->nfso_hash.le_prev = NULL;
4390 	newone = 0;
4391 	nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4392 	    lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4393 	ndp = dp;
4394 	if (NFSHASNFSV4N(nmp))
4395 		error = nfscl_tryopen(nmp, vp, lop->nfso_fh, lop->nfso_fhlen,
4396 		    lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4397 		    NULL, 0, &ndp, 0, 0, cred, p);
4398 	else
4399 		error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4400 		    np->n_v4->n4_fhlen, lop->nfso_fh, lop->nfso_fhlen,
4401 		    lop->nfso_mode, op, NFS4NODENAME(np->n_v4),
4402 		    np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4403 	if (error) {
4404 		if (newone)
4405 			nfscl_freeopen(op, 0, true);
4406 	} else {
4407 		op->nfso_mode |= lop->nfso_mode;
4408 		op->nfso_opencnt += lop->nfso_opencnt;
4409 		nfscl_freeopen(lop, 1, true);
4410 	}
4411 	if (nop != NULL)
4412 		free(nop, M_NFSCLOPEN);
4413 	if (ndp != NULL) {
4414 		/*
4415 		 * What should I do with the returned delegation, since the
4416 		 * delegation is being recalled? For now, just printf and
4417 		 * through it away.
4418 		 */
4419 		printf("Moveopen returned deleg\n");
4420 		free(ndp, M_NFSCLDELEG);
4421 	}
4422 	return (error);
4423 }
4424 
4425 /*
4426  * Recall all delegations on this client.
4427  */
4428 static void
4429 nfscl_totalrecall(struct nfsclclient *clp)
4430 {
4431 	struct nfscldeleg *dp;
4432 
4433 	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4434 		if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4435 			dp->nfsdl_flags |= NFSCLDL_RECALL;
4436 	}
4437 }
4438 
4439 /*
4440  * Relock byte ranges. Called for delegation recall and state expiry.
4441  */
4442 static int
4443 nfscl_relock(vnode_t vp, struct nfsclclient *clp, struct nfsmount *nmp,
4444     struct nfscllockowner *lp, struct nfscllock *lop, struct ucred *cred,
4445     NFSPROC_T *p)
4446 {
4447 	struct nfscllockowner *nlp;
4448 	struct nfsfh *nfhp;
4449 	struct nfsnode *np;
4450 	u_int64_t off, len;
4451 	int error, newone, donelocally;
4452 
4453 	if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) {
4454 		np = VTONFS(vp);
4455 		NFSLOCKNODE(np);
4456 		np->n_flag |= NMIGHTBELOCKED;
4457 		NFSUNLOCKNODE(np);
4458 	}
4459 
4460 	off = lop->nfslo_first;
4461 	len = lop->nfslo_end - lop->nfslo_first;
4462 	error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4463 	    clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4464 	    lp->nfsl_openowner, &nlp, &newone, &donelocally);
4465 	if (error || donelocally)
4466 		return (error);
4467 	nfhp = VTONFS(vp)->n_fhp;
4468 	error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4469 	    nfhp->nfh_len, nlp, newone, 0, off,
4470 	    len, lop->nfslo_type, cred, p);
4471 	if (error)
4472 		nfscl_freelockowner(nlp, 0);
4473 	return (error);
4474 }
4475 
4476 /*
4477  * Called to re-open a file. Basically get a vnode for the file handle
4478  * and then call nfsrpc_openrpc() to do the rest.
4479  */
4480 static int
4481 nfsrpc_reopen(struct nfsmount *nmp, u_int8_t *fhp, int fhlen,
4482     u_int32_t mode, struct nfsclopen *op, struct nfscldeleg **dpp,
4483     struct ucred *cred, NFSPROC_T *p)
4484 {
4485 	struct nfsnode *np;
4486 	vnode_t vp;
4487 	int error;
4488 
4489 	error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4490 	if (error)
4491 		return (error);
4492 	vp = NFSTOV(np);
4493 	if (NFSHASNFSV4N(nmp))
4494 		error = nfscl_tryopen(nmp, vp, fhp, fhlen, fhp, fhlen, mode, op,
4495 		    NULL, 0, dpp, 0, 0, cred, p);
4496 	else if (np->n_v4 != NULL)
4497 		error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4498 		    np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4499 		    NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4500 		    cred, p);
4501 	else
4502 		error = EINVAL;
4503 	vrele(vp);
4504 	return (error);
4505 }
4506 
4507 /*
4508  * Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4509  * NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4510  * fail.
4511  */
4512 static int
4513 nfscl_tryopen(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
4514     u_int8_t *newfhp, int newfhlen, u_int32_t mode, struct nfsclopen *op,
4515     u_int8_t *name, int namelen, struct nfscldeleg **ndpp,
4516     int reclaim, u_int32_t delegtype, struct ucred *cred, NFSPROC_T *p)
4517 {
4518 	int error;
4519 	struct nfscldeleg *dp;
4520 
4521 	dp = *ndpp;
4522 	do {
4523 		*ndpp = dp;	/* *ndpp needs to be set for retries. */
4524 		error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4525 		    mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4526 		    0, 0);
4527 		if (error == NFSERR_DELAY)
4528 			(void) nfs_catnap(PZERO, error, "nfstryop");
4529 	} while (error == NFSERR_DELAY);
4530 	if (error == EAUTH || error == EACCES) {
4531 		/* Try again using system credentials */
4532 		newnfs_setroot(cred);
4533 		do {
4534 		    *ndpp = dp;	/* *ndpp needs to be set for retries. */
4535 		    error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4536 			newfhlen, mode, op, name, namelen, ndpp, reclaim,
4537 			delegtype, cred, p, 1, 0);
4538 		    if (error == NFSERR_DELAY)
4539 			(void) nfs_catnap(PZERO, error, "nfstryop");
4540 		} while (error == NFSERR_DELAY);
4541 	}
4542 	return (error);
4543 }
4544 
4545 /*
4546  * Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4547  * NFSERR_DELAY. Also, retry with system credentials, if the provided
4548  * cred don't work.
4549  */
4550 static int
4551 nfscl_trylock(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp,
4552     int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4553     u_int64_t off, u_int64_t len, short type, struct ucred *cred, NFSPROC_T *p)
4554 {
4555 	struct nfsrv_descript nfsd, *nd = &nfsd;
4556 	int error;
4557 
4558 	do {
4559 		error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4560 		    reclaim, off, len, type, cred, p, 0);
4561 		if (!error && nd->nd_repstat == NFSERR_DELAY)
4562 			(void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4563 			    "nfstrylck");
4564 	} while (!error && nd->nd_repstat == NFSERR_DELAY);
4565 	if (!error)
4566 		error = nd->nd_repstat;
4567 	if (error == EAUTH || error == EACCES) {
4568 		/* Try again using root credentials */
4569 		newnfs_setroot(cred);
4570 		do {
4571 			error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4572 			    newone, reclaim, off, len, type, cred, p, 1);
4573 			if (!error && nd->nd_repstat == NFSERR_DELAY)
4574 				(void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4575 				    "nfstrylck");
4576 		} while (!error && nd->nd_repstat == NFSERR_DELAY);
4577 		if (!error)
4578 			error = nd->nd_repstat;
4579 	}
4580 	return (error);
4581 }
4582 
4583 /*
4584  * Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4585  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4586  * credentials fail.
4587  */
4588 int
4589 nfscl_trydelegreturn(struct nfscldeleg *dp, struct ucred *cred,
4590     struct nfsmount *nmp, NFSPROC_T *p)
4591 {
4592 	int error;
4593 
4594 	do {
4595 		error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4596 		if (error == NFSERR_DELAY)
4597 			(void) nfs_catnap(PZERO, error, "nfstrydp");
4598 	} while (error == NFSERR_DELAY);
4599 	if (error == EAUTH || error == EACCES) {
4600 		/* Try again using system credentials */
4601 		newnfs_setroot(cred);
4602 		do {
4603 			error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4604 			if (error == NFSERR_DELAY)
4605 				(void) nfs_catnap(PZERO, error, "nfstrydp");
4606 		} while (error == NFSERR_DELAY);
4607 	}
4608 	return (error);
4609 }
4610 
4611 /*
4612  * Try a close against the server. Just call nfsrpc_closerpc(),
4613  * retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4614  * credentials fail.
4615  */
4616 int
4617 nfscl_tryclose(struct nfsclopen *op, struct ucred *cred,
4618     struct nfsmount *nmp, NFSPROC_T *p, bool loop_on_delayed)
4619 {
4620 	struct nfsrv_descript nfsd, *nd = &nfsd;
4621 	int error;
4622 
4623 	do {
4624 		error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4625 		if (loop_on_delayed && error == NFSERR_DELAY)
4626 			(void) nfs_catnap(PZERO, error, "nfstrycl");
4627 	} while (loop_on_delayed && error == NFSERR_DELAY);
4628 	if (error == EAUTH || error == EACCES) {
4629 		/* Try again using system credentials */
4630 		newnfs_setroot(cred);
4631 		do {
4632 			error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4633 			if (loop_on_delayed && error == NFSERR_DELAY)
4634 				(void) nfs_catnap(PZERO, error, "nfstrycl");
4635 		} while (loop_on_delayed && error == NFSERR_DELAY);
4636 	}
4637 	return (error);
4638 }
4639 
4640 /*
4641  * Decide if a delegation on a file permits close without flushing writes
4642  * to the server. This might be a big performance win in some environments.
4643  * (Not useful until the client does caching on local stable storage.)
4644  */
4645 int
4646 nfscl_mustflush(vnode_t vp)
4647 {
4648 	struct nfsclclient *clp;
4649 	struct nfscldeleg *dp;
4650 	struct nfsnode *np;
4651 	struct nfsmount *nmp;
4652 
4653 	np = VTONFS(vp);
4654 	nmp = VFSTONFS(vp->v_mount);
4655 	if (!NFSHASNFSV4(nmp) || vp->v_type != VREG)
4656 		return (1);
4657 	NFSLOCKMNT(nmp);
4658 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4659 		NFSUNLOCKMNT(nmp);
4660 		return (1);
4661 	}
4662 	NFSUNLOCKMNT(nmp);
4663 	NFSLOCKCLSTATE();
4664 	clp = nfscl_findcl(nmp);
4665 	if (clp == NULL) {
4666 		NFSUNLOCKCLSTATE();
4667 		return (1);
4668 	}
4669 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4670 	if (dp != NULL && (dp->nfsdl_flags &
4671 	    (NFSCLDL_WRITE | NFSCLDL_RECALL | NFSCLDL_DELEGRET)) ==
4672 	     NFSCLDL_WRITE &&
4673 	    (dp->nfsdl_sizelimit >= np->n_size ||
4674 	     !NFSHASSTRICT3530(nmp))) {
4675 		NFSUNLOCKCLSTATE();
4676 		return (0);
4677 	}
4678 	NFSUNLOCKCLSTATE();
4679 	return (1);
4680 }
4681 
4682 /*
4683  * See if a (write) delegation exists for this file.
4684  */
4685 int
4686 nfscl_nodeleg(vnode_t vp, int writedeleg)
4687 {
4688 	struct nfsclclient *clp;
4689 	struct nfscldeleg *dp;
4690 	struct nfsnode *np;
4691 	struct nfsmount *nmp;
4692 
4693 	np = VTONFS(vp);
4694 	nmp = VFSTONFS(vp->v_mount);
4695 	if (!NFSHASNFSV4(nmp) || vp->v_type != VREG)
4696 		return (1);
4697 	NFSLOCKMNT(nmp);
4698 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4699 		NFSUNLOCKMNT(nmp);
4700 		return (1);
4701 	}
4702 	NFSUNLOCKMNT(nmp);
4703 	NFSLOCKCLSTATE();
4704 	clp = nfscl_findcl(nmp);
4705 	if (clp == NULL) {
4706 		NFSUNLOCKCLSTATE();
4707 		return (1);
4708 	}
4709 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4710 	if (dp != NULL &&
4711 	    (dp->nfsdl_flags & (NFSCLDL_RECALL | NFSCLDL_DELEGRET)) == 0 &&
4712 	    (writedeleg == 0 || (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4713 	     NFSCLDL_WRITE)) {
4714 		NFSUNLOCKCLSTATE();
4715 		return (0);
4716 	}
4717 	NFSUNLOCKCLSTATE();
4718 	return (1);
4719 }
4720 
4721 /*
4722  * Look for an associated delegation that should be DelegReturned.
4723  */
4724 int
4725 nfscl_removedeleg(vnode_t vp, NFSPROC_T *p, nfsv4stateid_t *stp)
4726 {
4727 	struct nfsclclient *clp;
4728 	struct nfscldeleg *dp;
4729 	struct nfsclowner *owp;
4730 	struct nfscllockowner *lp;
4731 	struct nfsmount *nmp;
4732 	struct mount *mp;
4733 	struct ucred *cred;
4734 	struct nfsnode *np;
4735 	int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4736 
4737 	nmp = VFSTONFS(vp->v_mount);
4738 	if (NFSHASPNFS(nmp))
4739 		return (retcnt);
4740 	NFSLOCKMNT(nmp);
4741 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4742 		NFSUNLOCKMNT(nmp);
4743 		return (retcnt);
4744 	}
4745 	NFSUNLOCKMNT(nmp);
4746 	np = VTONFS(vp);
4747 	mp = nmp->nm_mountp;
4748 	NFSLOCKCLSTATE();
4749 	/*
4750 	 * Loop around waiting for:
4751 	 * - outstanding I/O operations on delegations to complete
4752 	 * - for a delegation on vp that has state, lock the client and
4753 	 *   do a recall
4754 	 * - return delegation with no state
4755 	 */
4756 	while (1) {
4757 		clp = nfscl_findcl(nmp);
4758 		if (clp == NULL) {
4759 			NFSUNLOCKCLSTATE();
4760 			return (retcnt);
4761 		}
4762 		dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4763 		    np->n_fhp->nfh_len);
4764 		if (dp != NULL) {
4765 		    /*
4766 		     * Wait for outstanding I/O ops to be done.
4767 		     */
4768 		    if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4769 			if (igotlock) {
4770 			    nfsv4_unlock(&clp->nfsc_lock, 0);
4771 			    igotlock = 0;
4772 			}
4773 			dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4774 			msleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO,
4775 			    "nfscld", hz);
4776 			if (NFSCL_FORCEDISM(mp)) {
4777 			    dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
4778 			    NFSUNLOCKCLSTATE();
4779 			    return (0);
4780 			}
4781 			continue;
4782 		    }
4783 		    needsrecall = 0;
4784 		    LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4785 			if (!LIST_EMPTY(&owp->nfsow_open)) {
4786 			    needsrecall = 1;
4787 			    break;
4788 			}
4789 		    }
4790 		    if (!needsrecall) {
4791 			LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4792 			    if (!LIST_EMPTY(&lp->nfsl_lock)) {
4793 				needsrecall = 1;
4794 				break;
4795 			    }
4796 			}
4797 		    }
4798 		    if (needsrecall && !triedrecall) {
4799 			dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4800 			islept = 0;
4801 			while (!igotlock) {
4802 			    igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4803 				&islept, NFSCLSTATEMUTEXPTR, mp);
4804 			    if (NFSCL_FORCEDISM(mp)) {
4805 				dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
4806 				if (igotlock)
4807 				    nfsv4_unlock(&clp->nfsc_lock, 0);
4808 				NFSUNLOCKCLSTATE();
4809 				return (0);
4810 			    }
4811 			    if (islept)
4812 				break;
4813 			}
4814 			if (islept)
4815 			    continue;
4816 			NFSUNLOCKCLSTATE();
4817 			cred = newnfs_getcred();
4818 			newnfs_copycred(&dp->nfsdl_cred, cred);
4819 			nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0, NULL);
4820 			NFSFREECRED(cred);
4821 			triedrecall = 1;
4822 			NFSLOCKCLSTATE();
4823 			nfsv4_unlock(&clp->nfsc_lock, 0);
4824 			igotlock = 0;
4825 			continue;
4826 		    }
4827 		    *stp = dp->nfsdl_stateid;
4828 		    retcnt = 1;
4829 		    nfscl_cleandeleg(dp);
4830 		    nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4831 		}
4832 		if (igotlock)
4833 		    nfsv4_unlock(&clp->nfsc_lock, 0);
4834 		NFSUNLOCKCLSTATE();
4835 		return (retcnt);
4836 	}
4837 }
4838 
4839 /*
4840  * Look for associated delegation(s) that should be DelegReturned.
4841  */
4842 int
4843 nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t *fstp, int *gotfdp, vnode_t tvp,
4844     nfsv4stateid_t *tstp, int *gottdp, NFSPROC_T *p)
4845 {
4846 	struct nfsclclient *clp;
4847 	struct nfscldeleg *dp;
4848 	struct nfsclowner *owp;
4849 	struct nfscllockowner *lp;
4850 	struct nfsmount *nmp;
4851 	struct mount *mp;
4852 	struct ucred *cred;
4853 	struct nfsnode *np;
4854 	int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4855 
4856 	nmp = VFSTONFS(fvp->v_mount);
4857 	*gotfdp = 0;
4858 	*gottdp = 0;
4859 	if (NFSHASPNFS(nmp))
4860 		return (retcnt);
4861 	NFSLOCKMNT(nmp);
4862 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
4863 		NFSUNLOCKMNT(nmp);
4864 		return (retcnt);
4865 	}
4866 	NFSUNLOCKMNT(nmp);
4867 	mp = nmp->nm_mountp;
4868 	NFSLOCKCLSTATE();
4869 	/*
4870 	 * Loop around waiting for:
4871 	 * - outstanding I/O operations on delegations to complete
4872 	 * - for a delegation on fvp that has state, lock the client and
4873 	 *   do a recall
4874 	 * - return delegation(s) with no state.
4875 	 */
4876 	while (1) {
4877 		clp = nfscl_findcl(nmp);
4878 		if (clp == NULL) {
4879 			NFSUNLOCKCLSTATE();
4880 			return (retcnt);
4881 		}
4882 		np = VTONFS(fvp);
4883 		dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4884 		    np->n_fhp->nfh_len);
4885 		if (dp != NULL && *gotfdp == 0) {
4886 		    /*
4887 		     * Wait for outstanding I/O ops to be done.
4888 		     */
4889 		    if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4890 			if (igotlock) {
4891 			    nfsv4_unlock(&clp->nfsc_lock, 0);
4892 			    igotlock = 0;
4893 			}
4894 			dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4895 			msleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO,
4896 			    "nfscld", hz);
4897 			if (NFSCL_FORCEDISM(mp)) {
4898 			    dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
4899 			    NFSUNLOCKCLSTATE();
4900 			    *gotfdp = 0;
4901 			    *gottdp = 0;
4902 			    return (0);
4903 			}
4904 			continue;
4905 		    }
4906 		    needsrecall = 0;
4907 		    LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4908 			if (!LIST_EMPTY(&owp->nfsow_open)) {
4909 			    needsrecall = 1;
4910 			    break;
4911 			}
4912 		    }
4913 		    if (!needsrecall) {
4914 			LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4915 			    if (!LIST_EMPTY(&lp->nfsl_lock)) {
4916 				needsrecall = 1;
4917 				break;
4918 			    }
4919 			}
4920 		    }
4921 		    if (needsrecall && !triedrecall) {
4922 			dp->nfsdl_flags |= NFSCLDL_DELEGRET;
4923 			islept = 0;
4924 			while (!igotlock) {
4925 			    igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4926 				&islept, NFSCLSTATEMUTEXPTR, mp);
4927 			    if (NFSCL_FORCEDISM(mp)) {
4928 				dp->nfsdl_flags &= ~NFSCLDL_DELEGRET;
4929 				if (igotlock)
4930 				    nfsv4_unlock(&clp->nfsc_lock, 0);
4931 				NFSUNLOCKCLSTATE();
4932 				*gotfdp = 0;
4933 				*gottdp = 0;
4934 				return (0);
4935 			    }
4936 			    if (islept)
4937 				break;
4938 			}
4939 			if (islept)
4940 			    continue;
4941 			NFSUNLOCKCLSTATE();
4942 			cred = newnfs_getcred();
4943 			newnfs_copycred(&dp->nfsdl_cred, cred);
4944 			nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0, NULL);
4945 			NFSFREECRED(cred);
4946 			triedrecall = 1;
4947 			NFSLOCKCLSTATE();
4948 			nfsv4_unlock(&clp->nfsc_lock, 0);
4949 			igotlock = 0;
4950 			continue;
4951 		    }
4952 		    *fstp = dp->nfsdl_stateid;
4953 		    retcnt++;
4954 		    *gotfdp = 1;
4955 		    nfscl_cleandeleg(dp);
4956 		    nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4957 		}
4958 		if (igotlock) {
4959 		    nfsv4_unlock(&clp->nfsc_lock, 0);
4960 		    igotlock = 0;
4961 		}
4962 		if (tvp != NULL) {
4963 		    np = VTONFS(tvp);
4964 		    dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4965 			np->n_fhp->nfh_len);
4966 		    if (dp != NULL && *gottdp == 0) {
4967 			/*
4968 			 * Wait for outstanding I/O ops to be done.
4969 			 */
4970 			if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4971 			    dp->nfsdl_rwlock.nfslock_lock |= NFSV4LOCK_WANTED;
4972 			    msleep(&dp->nfsdl_rwlock, NFSCLSTATEMUTEXPTR, PZERO,
4973 				"nfscld", hz);
4974 			    if (NFSCL_FORCEDISM(mp)) {
4975 				NFSUNLOCKCLSTATE();
4976 				*gotfdp = 0;
4977 				*gottdp = 0;
4978 				return (0);
4979 			    }
4980 			    continue;
4981 			}
4982 			LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4983 			    if (!LIST_EMPTY(&owp->nfsow_open)) {
4984 				NFSUNLOCKCLSTATE();
4985 				return (retcnt);
4986 			    }
4987 			}
4988 			LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4989 			    if (!LIST_EMPTY(&lp->nfsl_lock)) {
4990 				NFSUNLOCKCLSTATE();
4991 				return (retcnt);
4992 			    }
4993 			}
4994 			*tstp = dp->nfsdl_stateid;
4995 			retcnt++;
4996 			*gottdp = 1;
4997 			nfscl_cleandeleg(dp);
4998 			nfscl_freedeleg(&clp->nfsc_deleg, dp, true);
4999 		    }
5000 		}
5001 		NFSUNLOCKCLSTATE();
5002 		return (retcnt);
5003 	}
5004 }
5005 
5006 /*
5007  * Get a reference on the clientid associated with the mount point.
5008  * Return 1 if success, 0 otherwise.
5009  */
5010 int
5011 nfscl_getref(struct nfsmount *nmp)
5012 {
5013 	struct nfsclclient *clp;
5014 	int ret;
5015 
5016 	NFSLOCKCLSTATE();
5017 	clp = nfscl_findcl(nmp);
5018 	if (clp == NULL) {
5019 		NFSUNLOCKCLSTATE();
5020 		return (0);
5021 	}
5022 	nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, nmp->nm_mountp);
5023 	ret = 1;
5024 	if (NFSCL_FORCEDISM(nmp->nm_mountp))
5025 		ret = 0;
5026 	NFSUNLOCKCLSTATE();
5027 	return (ret);
5028 }
5029 
5030 /*
5031  * Release a reference on a clientid acquired with the above call.
5032  */
5033 void
5034 nfscl_relref(struct nfsmount *nmp)
5035 {
5036 	struct nfsclclient *clp;
5037 
5038 	NFSLOCKCLSTATE();
5039 	clp = nfscl_findcl(nmp);
5040 	if (clp == NULL) {
5041 		NFSUNLOCKCLSTATE();
5042 		return;
5043 	}
5044 	nfsv4_relref(&clp->nfsc_lock);
5045 	NFSUNLOCKCLSTATE();
5046 }
5047 
5048 /*
5049  * Save the size attribute in the delegation, since the nfsnode
5050  * is going away.
5051  */
5052 void
5053 nfscl_reclaimnode(vnode_t vp)
5054 {
5055 	struct nfsclclient *clp;
5056 	struct nfscldeleg *dp;
5057 	struct nfsnode *np = VTONFS(vp);
5058 	struct nfsmount *nmp;
5059 
5060 	nmp = VFSTONFS(vp->v_mount);
5061 	if (!NFSHASNFSV4(nmp))
5062 		return;
5063 	NFSLOCKCLSTATE();
5064 	clp = nfscl_findcl(nmp);
5065 	if (clp == NULL) {
5066 		NFSUNLOCKCLSTATE();
5067 		return;
5068 	}
5069 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5070 	if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
5071 		dp->nfsdl_size = np->n_size;
5072 	NFSUNLOCKCLSTATE();
5073 }
5074 
5075 /*
5076  * Get the saved size attribute in the delegation, since it is a
5077  * newly allocated nfsnode.
5078  */
5079 void
5080 nfscl_newnode(vnode_t vp)
5081 {
5082 	struct nfsclclient *clp;
5083 	struct nfscldeleg *dp;
5084 	struct nfsnode *np = VTONFS(vp);
5085 	struct nfsmount *nmp;
5086 
5087 	nmp = VFSTONFS(vp->v_mount);
5088 	if (!NFSHASNFSV4(nmp))
5089 		return;
5090 	NFSLOCKCLSTATE();
5091 	clp = nfscl_findcl(nmp);
5092 	if (clp == NULL) {
5093 		NFSUNLOCKCLSTATE();
5094 		return;
5095 	}
5096 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5097 	if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
5098 		np->n_size = dp->nfsdl_size;
5099 	NFSUNLOCKCLSTATE();
5100 }
5101 
5102 /*
5103  * If there is a valid write delegation for this file, set the modtime
5104  * to the local clock time.
5105  */
5106 void
5107 nfscl_delegmodtime(vnode_t vp)
5108 {
5109 	struct nfsclclient *clp;
5110 	struct nfscldeleg *dp;
5111 	struct nfsnode *np = VTONFS(vp);
5112 	struct nfsmount *nmp;
5113 
5114 	nmp = VFSTONFS(vp->v_mount);
5115 	if (!NFSHASNFSV4(nmp))
5116 		return;
5117 	NFSLOCKMNT(nmp);
5118 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
5119 		NFSUNLOCKMNT(nmp);
5120 		return;
5121 	}
5122 	NFSUNLOCKMNT(nmp);
5123 	NFSLOCKCLSTATE();
5124 	clp = nfscl_findcl(nmp);
5125 	if (clp == NULL) {
5126 		NFSUNLOCKCLSTATE();
5127 		return;
5128 	}
5129 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5130 	if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
5131 		nanotime(&dp->nfsdl_modtime);
5132 		dp->nfsdl_flags |= NFSCLDL_MODTIMESET;
5133 	}
5134 	NFSUNLOCKCLSTATE();
5135 }
5136 
5137 /*
5138  * If there is a valid write delegation for this file with a modtime set,
5139  * put that modtime in mtime.
5140  */
5141 void
5142 nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
5143 {
5144 	struct nfsclclient *clp;
5145 	struct nfscldeleg *dp;
5146 	struct nfsnode *np = VTONFS(vp);
5147 	struct nfsmount *nmp;
5148 
5149 	nmp = VFSTONFS(vp->v_mount);
5150 	if (!NFSHASNFSV4(nmp))
5151 		return;
5152 	NFSLOCKMNT(nmp);
5153 	if ((nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0) {
5154 		NFSUNLOCKMNT(nmp);
5155 		return;
5156 	}
5157 	NFSUNLOCKMNT(nmp);
5158 	NFSLOCKCLSTATE();
5159 	clp = nfscl_findcl(nmp);
5160 	if (clp == NULL) {
5161 		NFSUNLOCKCLSTATE();
5162 		return;
5163 	}
5164 	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5165 	if (dp != NULL &&
5166 	    (dp->nfsdl_flags & (NFSCLDL_WRITE | NFSCLDL_MODTIMESET)) ==
5167 	    (NFSCLDL_WRITE | NFSCLDL_MODTIMESET))
5168 		*mtime = dp->nfsdl_modtime;
5169 	NFSUNLOCKCLSTATE();
5170 }
5171 
5172 static int
5173 nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
5174 {
5175 	short *defaulterrp, *errp;
5176 
5177 	if (!nd->nd_repstat)
5178 		return (0);
5179 	if (nd->nd_procnum == NFSPROC_NOOP)
5180 		return (txdr_unsigned(nd->nd_repstat & 0xffff));
5181 	if (nd->nd_repstat == EBADRPC)
5182 		return (txdr_unsigned(NFSERR_BADXDR));
5183 	if (nd->nd_repstat == NFSERR_MINORVERMISMATCH ||
5184 	    nd->nd_repstat == NFSERR_OPILLEGAL)
5185 		return (txdr_unsigned(nd->nd_repstat));
5186 	if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
5187 	    minorvers > NFSV4_MINORVERSION) {
5188 		/* NFSv4.n error. */
5189 		return (txdr_unsigned(nd->nd_repstat));
5190 	}
5191 	if (nd->nd_procnum < NFSV4OP_CBNOPS)
5192 		errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
5193 	else
5194 		return (txdr_unsigned(nd->nd_repstat));
5195 	while (*++errp)
5196 		if (*errp == (short)nd->nd_repstat)
5197 			return (txdr_unsigned(nd->nd_repstat));
5198 	return (txdr_unsigned(*defaulterrp));
5199 }
5200 
5201 /*
5202  * Called to find/add a layout to a client.
5203  * This function returns the layout with a refcnt (shared lock) upon
5204  * success (returns 0) or with no lock/refcnt on the layout when an
5205  * error is returned.
5206  * If a layout is passed in via lypp, it is locked (exclusively locked).
5207  */
5208 int
5209 nfscl_layout(struct nfsmount *nmp, vnode_t vp, u_int8_t *fhp, int fhlen,
5210     nfsv4stateid_t *stateidp, int layouttype, int retonclose,
5211     struct nfsclflayouthead *fhlp, struct nfscllayout **lypp,
5212     struct ucred *cred, NFSPROC_T *p)
5213 {
5214 	struct nfsclclient *clp;
5215 	struct nfscllayout *lyp, *tlyp;
5216 	struct nfsclflayout *flp;
5217 	struct nfsnode *np = VTONFS(vp);
5218 	mount_t mp;
5219 	int layout_passed_in;
5220 
5221 	mp = nmp->nm_mountp;
5222 	layout_passed_in = 1;
5223 	tlyp = NULL;
5224 	lyp = *lypp;
5225 	if (lyp == NULL) {
5226 		layout_passed_in = 0;
5227 		tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
5228 		    M_WAITOK | M_ZERO);
5229 	}
5230 
5231 	NFSLOCKCLSTATE();
5232 	clp = nmp->nm_clp;
5233 	if (clp == NULL) {
5234 		if (layout_passed_in != 0)
5235 			nfsv4_unlock(&lyp->nfsly_lock, 0);
5236 		NFSUNLOCKCLSTATE();
5237 		if (tlyp != NULL)
5238 			free(tlyp, M_NFSLAYOUT);
5239 		return (EPERM);
5240 	}
5241 	if (lyp == NULL) {
5242 		/*
5243 		 * Although no lyp was passed in, another thread might have
5244 		 * allocated one. If one is found, just increment it's ref
5245 		 * count and return it.
5246 		 */
5247 		lyp = nfscl_findlayout(clp, fhp, fhlen);
5248 		if (lyp == NULL) {
5249 			lyp = tlyp;
5250 			tlyp = NULL;
5251 			lyp->nfsly_stateid.seqid = stateidp->seqid;
5252 			lyp->nfsly_stateid.other[0] = stateidp->other[0];
5253 			lyp->nfsly_stateid.other[1] = stateidp->other[1];
5254 			lyp->nfsly_stateid.other[2] = stateidp->other[2];
5255 			lyp->nfsly_lastbyte = 0;
5256 			LIST_INIT(&lyp->nfsly_flayread);
5257 			LIST_INIT(&lyp->nfsly_flayrw);
5258 			LIST_INIT(&lyp->nfsly_recall);
5259 			lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
5260 			lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
5261 			lyp->nfsly_clp = clp;
5262 			if (layouttype == NFSLAYOUT_FLEXFILE)
5263 				lyp->nfsly_flags = NFSLY_FLEXFILE;
5264 			else
5265 				lyp->nfsly_flags = NFSLY_FILES;
5266 			if (retonclose != 0)
5267 				lyp->nfsly_flags |= NFSLY_RETONCLOSE;
5268 			lyp->nfsly_fhlen = fhlen;
5269 			NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
5270 			TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5271 			LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
5272 			    nfsly_hash);
5273 			lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5274 			nfscl_layoutcnt++;
5275 			nfsstatsv1.cllayouts++;
5276 		} else {
5277 			if (retonclose != 0)
5278 				lyp->nfsly_flags |= NFSLY_RETONCLOSE;
5279 			if (stateidp->seqid > lyp->nfsly_stateid.seqid)
5280 				lyp->nfsly_stateid.seqid = stateidp->seqid;
5281 			TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5282 			TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5283 			lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5284 		}
5285 		nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5286 		if (NFSCL_FORCEDISM(mp)) {
5287 			NFSUNLOCKCLSTATE();
5288 			if (tlyp != NULL)
5289 				free(tlyp, M_NFSLAYOUT);
5290 			return (EPERM);
5291 		}
5292 		*lypp = lyp;
5293 	} else if (stateidp->seqid > lyp->nfsly_stateid.seqid)
5294 		lyp->nfsly_stateid.seqid = stateidp->seqid;
5295 
5296 	/* Merge the new list of File Layouts into the list. */
5297 	flp = LIST_FIRST(fhlp);
5298 	if (flp != NULL) {
5299 		if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
5300 			nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
5301 		else
5302 			nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
5303 	}
5304 	if (layout_passed_in != 0)
5305 		nfsv4_unlock(&lyp->nfsly_lock, 1);
5306 	NFSUNLOCKCLSTATE();
5307 	if (tlyp != NULL)
5308 		free(tlyp, M_NFSLAYOUT);
5309 	return (0);
5310 }
5311 
5312 /*
5313  * Search for a layout by MDS file handle.
5314  * If one is found, it is returned with a refcnt (shared lock) iff
5315  * retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
5316  * returned NULL.
5317  */
5318 struct nfscllayout *
5319 nfscl_getlayout(struct nfsclclient *clp, uint8_t *fhp, int fhlen,
5320     uint64_t off, uint32_t rwaccess, struct nfsclflayout **retflpp,
5321     int *recalledp)
5322 {
5323 	struct nfscllayout *lyp;
5324 	mount_t mp;
5325 	int error, igotlock;
5326 
5327 	mp = clp->nfsc_nmp->nm_mountp;
5328 	*recalledp = 0;
5329 	*retflpp = NULL;
5330 	NFSLOCKCLSTATE();
5331 	lyp = nfscl_findlayout(clp, fhp, fhlen);
5332 	if (lyp != NULL) {
5333 		if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5334 			TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
5335 			TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
5336 			lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
5337 			error = nfscl_findlayoutforio(lyp, off, rwaccess,
5338 			    retflpp);
5339 			if (error == 0)
5340 				nfsv4_getref(&lyp->nfsly_lock, NULL,
5341 				    NFSCLSTATEMUTEXPTR, mp);
5342 			else {
5343 				do {
5344 					igotlock = nfsv4_lock(&lyp->nfsly_lock,
5345 					    1, NULL, NFSCLSTATEMUTEXPTR, mp);
5346 				} while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
5347 				*retflpp = NULL;
5348 			}
5349 			if (NFSCL_FORCEDISM(mp)) {
5350 				lyp = NULL;
5351 				*recalledp = 1;
5352 			}
5353 		} else {
5354 			lyp = NULL;
5355 			*recalledp = 1;
5356 		}
5357 	}
5358 	NFSUNLOCKCLSTATE();
5359 	return (lyp);
5360 }
5361 
5362 /*
5363  * Search for a layout by MDS file handle. If one is found, mark in to be
5364  * recalled, if it already marked "return on close".
5365  */
5366 static void
5367 nfscl_retoncloselayout(vnode_t vp, struct nfsclclient *clp, uint8_t *fhp,
5368     int fhlen, struct nfsclrecalllayout **recallpp, struct nfscllayout **lypp)
5369 {
5370 	struct nfscllayout *lyp;
5371 	uint32_t iomode;
5372 
5373 	*lypp = NULL;
5374 	if (vp->v_type != VREG || !NFSHASPNFS(VFSTONFS(vp->v_mount)) ||
5375 	    nfscl_enablecallb == 0 || nfs_numnfscbd == 0 ||
5376 	    (VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
5377 		return;
5378 	lyp = nfscl_findlayout(clp, fhp, fhlen);
5379 	if (lyp != NULL && (lyp->nfsly_flags & NFSLY_RETONCLOSE) != 0) {
5380 		if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5381 			iomode = 0;
5382 			if (!LIST_EMPTY(&lyp->nfsly_flayread))
5383 				iomode |= NFSLAYOUTIOMODE_READ;
5384 			if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5385 				iomode |= NFSLAYOUTIOMODE_RW;
5386 			nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5387 			    0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
5388 			    *recallpp);
5389 			NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
5390 			*recallpp = NULL;
5391 		}
5392 
5393 		/* Now, wake up renew thread to do LayoutReturn. */
5394 		wakeup(clp);
5395 		*lypp = lyp;
5396 	}
5397 }
5398 
5399 /*
5400  * Mark the layout to be recalled and with an error.
5401  * Also, disable the dsp from further use.
5402  */
5403 void
5404 nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
5405     struct nfscllayout *lyp, struct nfsclds *dsp)
5406 {
5407 	struct nfsclrecalllayout *recallp;
5408 	uint32_t iomode;
5409 
5410 	printf("DS being disabled, error=%d\n", stat);
5411 	/* Set up the return of the layout. */
5412 	recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
5413 	iomode = 0;
5414 	NFSLOCKCLSTATE();
5415 	if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
5416 		if (!LIST_EMPTY(&lyp->nfsly_flayread))
5417 			iomode |= NFSLAYOUTIOMODE_READ;
5418 		if (!LIST_EMPTY(&lyp->nfsly_flayrw))
5419 			iomode |= NFSLAYOUTIOMODE_RW;
5420 		(void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
5421 		    0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
5422 		    dp->nfsdi_deviceid, recallp);
5423 		NFSUNLOCKCLSTATE();
5424 		NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
5425 	} else {
5426 		NFSUNLOCKCLSTATE();
5427 		free(recallp, M_NFSLAYRECALL);
5428 	}
5429 
5430 	/* And shut the TCP connection down. */
5431 	nfscl_cancelreqs(dsp);
5432 }
5433 
5434 /*
5435  * Cancel all RPCs for this "dsp" by closing the connection.
5436  * Also, mark the session as defunct.
5437  * If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5438  * cannot be shut down.
5439  */
5440 void
5441 nfscl_cancelreqs(struct nfsclds *dsp)
5442 {
5443 	struct __rpc_client *cl;
5444 	static int non_event;
5445 
5446 	NFSLOCKDS(dsp);
5447 	if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED | NFSCLDS_SAMECONN)) == 0 &&
5448 	    dsp->nfsclds_sockp != NULL &&
5449 	    dsp->nfsclds_sockp->nr_client != NULL) {
5450 		dsp->nfsclds_flags |= NFSCLDS_CLOSED;
5451 		cl = dsp->nfsclds_sockp->nr_client;
5452 		dsp->nfsclds_sess.nfsess_defunct = 1;
5453 		NFSUNLOCKDS(dsp);
5454 		CLNT_CLOSE(cl);
5455 		/*
5456 		 * This 1sec sleep is done to reduce the number of reconnect
5457 		 * attempts made on the DS while it has failed.
5458 		 */
5459 		tsleep(&non_event, PVFS, "ndscls", hz);
5460 		return;
5461 	}
5462 	NFSUNLOCKDS(dsp);
5463 }
5464 
5465 /*
5466  * Dereference a layout.
5467  */
5468 void
5469 nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5470 {
5471 
5472 	NFSLOCKCLSTATE();
5473 	if (exclocked != 0)
5474 		nfsv4_unlock(&lyp->nfsly_lock, 0);
5475 	else
5476 		nfsv4_relref(&lyp->nfsly_lock);
5477 	NFSUNLOCKCLSTATE();
5478 }
5479 
5480 /*
5481  * Search for a devinfo by deviceid. If one is found, return it after
5482  * acquiring a reference count on it.
5483  */
5484 struct nfscldevinfo *
5485 nfscl_getdevinfo(struct nfsclclient *clp, uint8_t *deviceid,
5486     struct nfscldevinfo *dip)
5487 {
5488 
5489 	NFSLOCKCLSTATE();
5490 	if (dip == NULL)
5491 		dip = nfscl_finddevinfo(clp, deviceid);
5492 	if (dip != NULL)
5493 		dip->nfsdi_refcnt++;
5494 	NFSUNLOCKCLSTATE();
5495 	return (dip);
5496 }
5497 
5498 /*
5499  * Dereference a devinfo structure.
5500  */
5501 static void
5502 nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5503 {
5504 
5505 	dip->nfsdi_refcnt--;
5506 	if (dip->nfsdi_refcnt == 0)
5507 		wakeup(&dip->nfsdi_refcnt);
5508 }
5509 
5510 /*
5511  * Dereference a devinfo structure.
5512  */
5513 void
5514 nfscl_reldevinfo(struct nfscldevinfo *dip)
5515 {
5516 
5517 	NFSLOCKCLSTATE();
5518 	nfscl_reldevinfo_locked(dip);
5519 	NFSUNLOCKCLSTATE();
5520 }
5521 
5522 /*
5523  * Find a layout for this file handle. Return NULL upon failure.
5524  */
5525 static struct nfscllayout *
5526 nfscl_findlayout(struct nfsclclient *clp, u_int8_t *fhp, int fhlen)
5527 {
5528 	struct nfscllayout *lyp;
5529 
5530 	LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5531 		if (lyp->nfsly_fhlen == fhlen &&
5532 		    !NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5533 			break;
5534 	return (lyp);
5535 }
5536 
5537 /*
5538  * Find a devinfo for this deviceid. Return NULL upon failure.
5539  */
5540 static struct nfscldevinfo *
5541 nfscl_finddevinfo(struct nfsclclient *clp, uint8_t *deviceid)
5542 {
5543 	struct nfscldevinfo *dip;
5544 
5545 	LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5546 		if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5547 		    == 0)
5548 			break;
5549 	return (dip);
5550 }
5551 
5552 /*
5553  * Merge the new file layout list into the main one, maintaining it in
5554  * increasing offset order.
5555  */
5556 static void
5557 nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5558     struct nfsclflayouthead *newfhlp)
5559 {
5560 	struct nfsclflayout *flp, *nflp, *prevflp, *tflp;
5561 
5562 	flp = LIST_FIRST(fhlp);
5563 	prevflp = NULL;
5564 	LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5565 		while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5566 			prevflp = flp;
5567 			flp = LIST_NEXT(flp, nfsfl_list);
5568 		}
5569 		if (prevflp == NULL)
5570 			LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5571 		else
5572 			LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5573 		prevflp = nflp;
5574 	}
5575 }
5576 
5577 /*
5578  * Add this nfscldevinfo to the client, if it doesn't already exist.
5579  * This function consumes the structure pointed at by dip, if not NULL.
5580  */
5581 int
5582 nfscl_adddevinfo(struct nfsmount *nmp, struct nfscldevinfo *dip, int ind,
5583     struct nfsclflayout *flp)
5584 {
5585 	struct nfsclclient *clp;
5586 	struct nfscldevinfo *tdip;
5587 	uint8_t *dev;
5588 
5589 	NFSLOCKCLSTATE();
5590 	clp = nmp->nm_clp;
5591 	if (clp == NULL) {
5592 		NFSUNLOCKCLSTATE();
5593 		if (dip != NULL)
5594 			free(dip, M_NFSDEVINFO);
5595 		return (ENODEV);
5596 	}
5597 	if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5598 		dev = flp->nfsfl_dev;
5599 	else
5600 		dev = flp->nfsfl_ffm[ind].dev;
5601 	tdip = nfscl_finddevinfo(clp, dev);
5602 	if (tdip != NULL) {
5603 		tdip->nfsdi_layoutrefs++;
5604 		if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5605 			flp->nfsfl_devp = tdip;
5606 		else
5607 			flp->nfsfl_ffm[ind].devp = tdip;
5608 		nfscl_reldevinfo_locked(tdip);
5609 		NFSUNLOCKCLSTATE();
5610 		if (dip != NULL)
5611 			free(dip, M_NFSDEVINFO);
5612 		return (0);
5613 	}
5614 	if (dip != NULL) {
5615 		LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5616 		dip->nfsdi_layoutrefs = 1;
5617 		if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5618 			flp->nfsfl_devp = dip;
5619 		else
5620 			flp->nfsfl_ffm[ind].devp = dip;
5621 	}
5622 	NFSUNLOCKCLSTATE();
5623 	if (dip == NULL)
5624 		return (ENODEV);
5625 	return (0);
5626 }
5627 
5628 /*
5629  * Free up a layout structure and associated file layout structure(s).
5630  */
5631 void
5632 nfscl_freelayout(struct nfscllayout *layp)
5633 {
5634 	struct nfsclflayout *flp, *nflp;
5635 	struct nfsclrecalllayout *rp, *nrp;
5636 
5637 	LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5638 		LIST_REMOVE(flp, nfsfl_list);
5639 		nfscl_freeflayout(flp);
5640 	}
5641 	LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5642 		LIST_REMOVE(flp, nfsfl_list);
5643 		nfscl_freeflayout(flp);
5644 	}
5645 	LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5646 		LIST_REMOVE(rp, nfsrecly_list);
5647 		free(rp, M_NFSLAYRECALL);
5648 	}
5649 	nfscl_layoutcnt--;
5650 	nfsstatsv1.cllayouts--;
5651 	free(layp, M_NFSLAYOUT);
5652 }
5653 
5654 /*
5655  * Free up a file layout structure.
5656  */
5657 void
5658 nfscl_freeflayout(struct nfsclflayout *flp)
5659 {
5660 	int i, j;
5661 
5662 	if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5663 		for (i = 0; i < flp->nfsfl_fhcnt; i++)
5664 			free(flp->nfsfl_fh[i], M_NFSFH);
5665 		if (flp->nfsfl_devp != NULL)
5666 			flp->nfsfl_devp->nfsdi_layoutrefs--;
5667 	}
5668 	if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5669 		for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5670 			for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5671 				free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5672 			if (flp->nfsfl_ffm[i].devp != NULL)
5673 				flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5674 		}
5675 	free(flp, M_NFSFLAYOUT);
5676 }
5677 
5678 /*
5679  * Free up a file layout devinfo structure.
5680  */
5681 void
5682 nfscl_freedevinfo(struct nfscldevinfo *dip)
5683 {
5684 
5685 	free(dip, M_NFSDEVINFO);
5686 }
5687 
5688 /*
5689  * Mark any layouts that match as recalled.
5690  */
5691 static int
5692 nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5693     uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5694     char *devid, struct nfsclrecalllayout *recallp)
5695 {
5696 	struct nfsclrecalllayout *rp, *orp;
5697 
5698 	recallp->nfsrecly_recalltype = recalltype;
5699 	recallp->nfsrecly_iomode = iomode;
5700 	recallp->nfsrecly_stateseqid = stateseqid;
5701 	recallp->nfsrecly_off = off;
5702 	recallp->nfsrecly_len = len;
5703 	recallp->nfsrecly_stat = stat;
5704 	recallp->nfsrecly_op = op;
5705 	if (devid != NULL)
5706 		NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5707 	/*
5708 	 * Order the list as file returns first, followed by fsid and any
5709 	 * returns, both in increasing stateseqid order.
5710 	 * Note that the seqids wrap around, so 1 is after 0xffffffff.
5711 	 * (I'm not sure this is correct because I find RFC5661 confusing
5712 	 *  on this, but hopefully it will work ok.)
5713 	 */
5714 	orp = NULL;
5715 	LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5716 		orp = rp;
5717 		if ((recalltype == NFSLAYOUTRETURN_FILE &&
5718 		     (rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE ||
5719 		      nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) ||
5720 		    (recalltype != NFSLAYOUTRETURN_FILE &&
5721 		     rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5722 		     nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5723 			LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5724 			break;
5725 		}
5726 
5727 		/*
5728 		 * Put any error return on all the file returns that will
5729 		 * preceed this one.
5730 		 */
5731 		if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5732 		   stat != 0 && rp->nfsrecly_stat == 0) {
5733 			rp->nfsrecly_stat = stat;
5734 			rp->nfsrecly_op = op;
5735 			if (devid != NULL)
5736 				NFSBCOPY(devid, rp->nfsrecly_devid,
5737 				    NFSX_V4DEVICEID);
5738 		}
5739 	}
5740 	if (rp == NULL) {
5741 		if (orp == NULL)
5742 			LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5743 			    nfsrecly_list);
5744 		else
5745 			LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5746 	}
5747 	lyp->nfsly_flags |= NFSLY_RECALL;
5748 	wakeup(lyp->nfsly_clp);
5749 	return (0);
5750 }
5751 
5752 /*
5753  * Compare the two seqids for ordering. The trick is that the seqids can
5754  * wrap around from 0xffffffff->0, so check for the cases where one
5755  * has wrapped around.
5756  * Return 1 if seqid1 comes before seqid2, 0 otherwise.
5757  */
5758 static int
5759 nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5760 {
5761 
5762 	if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5763 		/* seqid2 has wrapped around. */
5764 		return (0);
5765 	if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5766 		/* seqid1 has wrapped around. */
5767 		return (1);
5768 	if (seqid1 <= seqid2)
5769 		return (1);
5770 	return (0);
5771 }
5772 
5773 /*
5774  * Do a layout return for each of the recalls.
5775  */
5776 static void
5777 nfscl_layoutreturn(struct nfsmount *nmp, struct nfscllayout *lyp,
5778     struct ucred *cred, NFSPROC_T *p)
5779 {
5780 	struct nfsclrecalllayout *rp;
5781 	nfsv4stateid_t stateid;
5782 	int layouttype;
5783 
5784 	NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5785 	stateid.seqid = lyp->nfsly_stateid.seqid;
5786 	if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5787 		layouttype = NFSLAYOUT_NFSV4_1_FILES;
5788 	else
5789 		layouttype = NFSLAYOUT_FLEXFILE;
5790 	LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5791 		(void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5792 		    lyp->nfsly_fhlen, 0, layouttype,
5793 		    rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5794 		    rp->nfsrecly_off, rp->nfsrecly_len,
5795 		    &stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5796 		    rp->nfsrecly_devid);
5797 	}
5798 }
5799 
5800 /*
5801  * Do the layout commit for a file layout.
5802  */
5803 static void
5804 nfscl_dolayoutcommit(struct nfsmount *nmp, struct nfscllayout *lyp,
5805     struct ucred *cred, NFSPROC_T *p)
5806 {
5807 	struct nfsclflayout *flp;
5808 	uint64_t len;
5809 	int error, layouttype;
5810 
5811 	if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5812 		layouttype = NFSLAYOUT_NFSV4_1_FILES;
5813 	else
5814 		layouttype = NFSLAYOUT_FLEXFILE;
5815 	LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5816 		if (layouttype == NFSLAYOUT_FLEXFILE &&
5817 		    (flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5818 			NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5819 			/* If not supported, don't bother doing it. */
5820 			NFSLOCKMNT(nmp);
5821 			nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5822 			NFSUNLOCKMNT(nmp);
5823 			break;
5824 		} else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5825 			len = flp->nfsfl_end - flp->nfsfl_off;
5826 			error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5827 			    lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5828 			    lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5829 			    layouttype, cred, p);
5830 			NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5831 			if (error == NFSERR_NOTSUPP) {
5832 				/* If not supported, don't bother doing it. */
5833 				NFSLOCKMNT(nmp);
5834 				nmp->nm_state |= NFSSTA_NOLAYOUTCOMMIT;
5835 				NFSUNLOCKMNT(nmp);
5836 				break;
5837 			}
5838 		}
5839 	}
5840 }
5841 
5842 /*
5843  * Commit all layouts for a file (vnode).
5844  */
5845 int
5846 nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5847 {
5848 	struct nfsclclient *clp;
5849 	struct nfscllayout *lyp;
5850 	struct nfsnode *np = VTONFS(vp);
5851 	mount_t mp;
5852 	struct nfsmount *nmp;
5853 
5854 	mp = vp->v_mount;
5855 	nmp = VFSTONFS(mp);
5856 	if (NFSHASNOLAYOUTCOMMIT(nmp))
5857 		return (0);
5858 	NFSLOCKCLSTATE();
5859 	clp = nmp->nm_clp;
5860 	if (clp == NULL) {
5861 		NFSUNLOCKCLSTATE();
5862 		return (EPERM);
5863 	}
5864 	lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5865 	if (lyp == NULL) {
5866 		NFSUNLOCKCLSTATE();
5867 		return (EPERM);
5868 	}
5869 	nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5870 	if (NFSCL_FORCEDISM(mp)) {
5871 		NFSUNLOCKCLSTATE();
5872 		return (EPERM);
5873 	}
5874 tryagain:
5875 	if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5876 		lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5877 		NFSUNLOCKCLSTATE();
5878 		NFSCL_DEBUG(4, "do layoutcommit2\n");
5879 		nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5880 		NFSLOCKCLSTATE();
5881 		goto tryagain;
5882 	}
5883 	nfsv4_relref(&lyp->nfsly_lock);
5884 	NFSUNLOCKCLSTATE();
5885 	return (0);
5886 }
5887