Context Navigation

source: rtems-libbsd/freebsd/sys/fs/nfsclient/nfs_clstate.c @ 6e4709b

6-freebsd-12

Last change on this file since 6e4709b was 6e4709b, checked in by Sebastian Huber <sebastian.huber@…>, on 01/05/23 at 16:42:48
vfs/nfs: Revert white space changes
Property mode set to `100644`
File size: 147.1 KB

Line
1	#include <machine/rtems-bsd-kernel-space.h>
2
3	/*-
4	* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
5	*
6	* Copyright (c) 2009 Rick Macklem, University of Guelph
7	* All rights reserved.
8	*
9	* Redistribution and use in source and binary forms, with or without
10	* modification, are permitted provided that the following conditions
11	* are met:
12	* 1. Redistributions of source code must retain the above copyright
13	* notice, this list of conditions and the following disclaimer.
14	* 2. Redistributions in binary form must reproduce the above copyright
15	* notice, this list of conditions and the following disclaimer in the
16	* documentation and/or other materials provided with the distribution.
17	*
18	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19	* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20	* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21	* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22	* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23	* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24	* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25	* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26	* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27	* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28	* SUCH DAMAGE.
29	*
30	*/
31
32	#include <sys/cdefs.h>
33	__FBSDID("$FreeBSD$");
34
35	/*
36	* These functions implement the client side state handling for NFSv4.
37	* NFSv4 state handling:
38	* - A lockowner is used to determine lock contention, so it
39	* corresponds directly to a Posix pid. (1 to 1 mapping)
40	* - The correct granularity of an OpenOwner is not nearly so
41	* obvious. An OpenOwner does the following:
42	* - provides a serial sequencing of Open/Close/Lock-with-new-lockowner
43	* - is used to check for Open/Share contention (not applicable to
44	* this client, since all Opens are Deny_None)
45	* As such, I considered both extreme.
46	* 1 OpenOwner per ClientID - Simple to manage, but fully serializes
47	* all Open, Close and Lock (with a new lockowner) Ops.
48	* 1 OpenOwner for each Open - This one results in an OpenConfirm for
49	* every Open, for most servers.
50	* So, I chose to use the same mapping as I did for LockOwnwers.
51	* The main concern here is that you can end up with multiple Opens
52	* for the same File Handle, but on different OpenOwners (opens
53	* inherited from parents, grandparents...) and you do not know
54	* which of these the vnodeop close applies to. This is handled by
55	* delaying the Close Op(s) until all of the Opens have been closed.
56	* (It is not yet obvious if this is the correct granularity.)
57	* - How the code handles serialization:
58	* - For the ClientId, it uses an exclusive lock while getting its
59	* SetClientId and during recovery. Otherwise, it uses a shared
60	* lock via a reference count.
61	* - For the rest of the data structures, it uses an SMP mutex
62	* (once the nfs client is SMP safe) and doesn't sleep while
63	* manipulating the linked lists.
64	* - The serialization of Open/Close/Lock/LockU falls out in the
65	* "wash", since OpenOwners and LockOwners are both mapped from
66	* Posix pid. In other words, there is only one Posix pid using
67	* any given owner, so that owner is serialized. (If you change
68	* the granularity of the OpenOwner, then code must be added to
69	* serialize Ops on the OpenOwner.)
70	* - When to get rid of OpenOwners and LockOwners.
71	* - The function nfscl_cleanup_common() is executed after a process exits.
72	* It goes through the client list looking for all Open and Lock Owners.
73	* When one is found, it is marked "defunct" or in the case of
74	* an OpenOwner without any Opens, freed.
75	* The renew thread scans for defunct Owners and gets rid of them,
76	* if it can. The LockOwners will also be deleted when the
77	* associated Open is closed.
78	* - If the LockU or Close Op(s) fail during close in a way
79	* that could be recovered upon retry, they are relinked to the
80	* ClientId's defunct open list and retried by the renew thread
81	* until they succeed or an unmount/recovery occurs.
82	* (Since we are done with them, they do not need to be recovered.)
83	*/
84
85	#ifndef APPLEKEXT
86	#include <fs/nfs/nfsport.h>
87
88	/*
89	* Global variables
90	*/
91	extern struct nfsstatsv1 nfsstatsv1;
92	extern struct nfsreqhead nfsd_reqq;
93	extern u_int32_t newnfs_false, newnfs_true;
94	extern int nfscl_debuglevel;
95	extern int nfscl_enablecallb;
96	extern int nfs_numnfscbd;
97	NFSREQSPINLOCK;
98	NFSCLSTATEMUTEX;
99	int nfscl_inited = 0;
100	struct nfsclhead nfsclhead; /* Head of clientid list */
101	int nfscl_deleghighwater = NFSCLDELEGHIGHWATER;
102	int nfscl_layouthighwater = NFSCLLAYOUTHIGHWATER;
103	#endif /* !APPLEKEXT */
104
105	static int nfscl_delegcnt = 0;
106	static int nfscl_layoutcnt = 0;
107	static int nfscl_getopen(struct nfsclownerhead , u_int8_t , int, u_int8_t *,
108	u_int8_t , u_int32_t, struct nfscllockowner , struct nfsclopen *);
109	static void nfscl_clrelease(struct nfsclclient *);
110	static void nfscl_cleanclient(struct nfsclclient *);
111	static void nfscl_expireclient(struct nfsclclient , struct nfsmount ,
112	struct ucred , NFSPROC_T );
113	static int nfscl_expireopen(struct nfsclclient , struct nfsclopen ,
114	struct nfsmount , struct ucred , NFSPROC_T *);
115	static void nfscl_recover(struct nfsclclient , struct ucred , NFSPROC_T *);
116	static void nfscl_insertlock(struct nfscllockowner , struct nfscllock ,
117	struct nfscllock *, int);
118	static int nfscl_updatelock(struct nfscllockowner , struct nfscllock *,
119	struct nfscllock **, int);
120	static void nfscl_delegreturnall(struct nfsclclient , NFSPROC_T );
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 **);
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	static int nfscl_errmap(struct nfsrv_descript *, u_int32_t);
157	static void nfscl_cleanup_common(struct nfsclclient , u_int8_t );
158	static int nfscl_recalldeleg(struct nfsclclient , struct nfsmount ,
159	struct nfscldeleg , vnode_t, struct ucred , NFSPROC_T *, int);
160	static void nfscl_freeopenowner(struct nfsclowner *, int);
161	static void nfscl_cleandeleg(struct nfscldeleg *);
162	static int nfscl_trydelegreturn(struct nfscldeleg , struct ucred ,
163	struct nfsmount , NFSPROC_T );
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	APPLESTATIC 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)
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	ret = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
245	if (ret != 0) {
246	free(nowp, M_NFSCLOWNER);
247	if (nop != NULL)
248	free(nop, M_NFSCLOPEN);
249	return (ret);
250	}
251
252	/*
253	* Get the Open iff it already exists.
254	* If none found, add the new one or return error, depending upon
255	* "create".
256	*/
257	NFSLOCKCLSTATE();
258	dp = NULL;
259	/* First check the delegation list */
260	if (nfhp != NULL && usedeleg) {
261	LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
262	if (dp->nfsdl_fhlen == fhlen &&
263	!NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
264	if (!(amode & NFSV4OPEN_ACCESSWRITE) \|\|
265	(dp->nfsdl_flags & NFSCLDL_WRITE))
266	break;
267	dp = NULL;
268	break;
269	}
270	}
271	}
272
273	if (dp != NULL) {
274	nfscl_filllockowner(p->td_proc, own, F_POSIX);
275	ohp = &dp->nfsdl_owner;
276	} else {
277	/* For NFSv4.1 and this option, use a single open_owner. */
278	if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
279	nfscl_filllockowner(NULL, own, F_POSIX);
280	else
281	nfscl_filllockowner(p->td_proc, own, F_POSIX);
282	ohp = &clp->nfsc_owner;
283	}
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(vnode_mount(vp)))) {
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	nfsstatsv1.clopens++;
416	}
417	LIST_INSERT_HEAD(&owp->nfsow_open, nop, nfso_list);
418	*opp = nop;
419	*nopp = NULL;
420	if (newonep != NULL)
421	*newonep = 1;
422	} else {
423	*opp = op;
424	}
425	}
426	}
427
428	/*
429	* Called to find/add a delegation to a client.
430	*/
431	APPLESTATIC int
432	nfscl_deleg(mount_t mp, struct nfsclclient clp, u_int8_t nfhp,
433	int fhlen, struct ucred cred, NFSPROC_T p, struct nfscldeleg **dpp)
434	{
435	struct nfscldeleg dp = dpp, *tdp;
436
437	/*
438	* First, if we have received a Read delegation for a file on a
439	* read/write file system, just return it, because they aren't
440	* useful, imho.
441	*/
442	if (mp != NULL && dp != NULL && !NFSMNT_RDONLY(mp) &&
443	(dp->nfsdl_flags & NFSCLDL_READ)) {
444	(void) nfscl_trydelegreturn(dp, cred, VFSTONFS(mp), p);
445	free(dp, M_NFSCLDELEG);
446	*dpp = NULL;
447	return (0);
448	}
449
450	/* Look for the correct deleg, based upon FH */
451	NFSLOCKCLSTATE();
452	tdp = nfscl_finddeleg(clp, nfhp, fhlen);
453	if (tdp == NULL) {
454	if (dp == NULL) {
455	NFSUNLOCKCLSTATE();
456	return (NFSERR_BADSTATEID);
457	}
458	*dpp = NULL;
459	TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
460	LIST_INSERT_HEAD(NFSCLDELEGHASH(clp, nfhp, fhlen), dp,
461	nfsdl_hash);
462	dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
463	nfsstatsv1.cldelegates++;
464	nfscl_delegcnt++;
465	} else {
466	/*
467	* Delegation already exists, what do we do if a new one??
468	*/
469	if (dp != NULL) {
470	printf("Deleg already exists!\n");
471	free(dp, M_NFSCLDELEG);
472	*dpp = NULL;
473	} else {
474	*dpp = tdp;
475	}
476	}
477	NFSUNLOCKCLSTATE();
478	return (0);
479	}
480
481	/*
482	* Find a delegation for this file handle. Return NULL upon failure.
483	*/
484	static struct nfscldeleg *
485	nfscl_finddeleg(struct nfsclclient clp, u_int8_t fhp, int fhlen)
486	{
487	struct nfscldeleg *dp;
488
489	LIST_FOREACH(dp, NFSCLDELEGHASH(clp, fhp, fhlen), nfsdl_hash) {
490	if (dp->nfsdl_fhlen == fhlen &&
491	!NFSBCMP(dp->nfsdl_fh, fhp, fhlen))
492	break;
493	}
494	return (dp);
495	}
496
497	/*
498	* Get a stateid for an I/O operation. First, look for an open and iff
499	* found, return either a lockowner stateid or the open stateid.
500	* If no Open is found, just return error and the special stateid of all zeros.
501	*/
502	APPLESTATIC int
503	nfscl_getstateid(vnode_t vp, u_int8_t *nfhp, int fhlen, u_int32_t mode,
504	int fords, struct ucred cred, NFSPROC_T p, nfsv4stateid_t *stateidp,
505	void **lckpp)
506	{
507	struct nfsclclient *clp;
508	struct nfsclowner *owp;
509	struct nfsclopen op = NULL, top;
510	struct nfscllockowner *lp;
511	struct nfscldeleg *dp;
512	struct nfsnode *np;
513	struct nfsmount *nmp;
514	u_int8_t own[NFSV4CL_LOCKNAMELEN];
515	int error, done;
516
517	*lckpp = NULL;
518	/*
519	* Initially, just set the special stateid of all zeros.
520	* (Don't do this for a DS, since the special stateid can't be used.)
521	*/
522	if (fords == 0) {
523	stateidp->seqid = 0;
524	stateidp->other[0] = 0;
525	stateidp->other[1] = 0;
526	stateidp->other[2] = 0;
527	}
528	if (vnode_vtype(vp) != VREG)
529	return (EISDIR);
530	np = VTONFS(vp);
531	nmp = VFSTONFS(vnode_mount(vp));
532	NFSLOCKCLSTATE();
533	clp = nfscl_findcl(nmp);
534	if (clp == NULL) {
535	NFSUNLOCKCLSTATE();
536	return (EACCES);
537	}
538
539	/*
540	* Wait for recovery to complete.
541	*/
542	while ((clp->nfsc_flags & NFSCLFLAGS_RECVRINPROG))
543	(void) nfsmsleep(&clp->nfsc_flags, NFSCLSTATEMUTEXPTR,
544	PZERO, "nfsrecvr", NULL);
545
546	/*
547	* First, look for a delegation.
548	*/
549	LIST_FOREACH(dp, NFSCLDELEGHASH(clp, nfhp, fhlen), nfsdl_hash) {
550	if (dp->nfsdl_fhlen == fhlen &&
551	!NFSBCMP(nfhp, dp->nfsdl_fh, fhlen)) {
552	if (!(mode & NFSV4OPEN_ACCESSWRITE) \|\|
553	(dp->nfsdl_flags & NFSCLDL_WRITE)) {
554	stateidp->seqid = dp->nfsdl_stateid.seqid;
555	stateidp->other[0] = dp->nfsdl_stateid.other[0];
556	stateidp->other[1] = dp->nfsdl_stateid.other[1];
557	stateidp->other[2] = dp->nfsdl_stateid.other[2];
558	if (!(np->n_flag & NDELEGRECALL)) {
559	TAILQ_REMOVE(&clp->nfsc_deleg, dp,
560	nfsdl_list);
561	TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp,
562	nfsdl_list);
563	dp->nfsdl_timestamp = NFSD_MONOSEC +
564	120;
565	dp->nfsdl_rwlock.nfslock_usecnt++;
566	lckpp = (void )&dp->nfsdl_rwlock;
567	}
568	NFSUNLOCKCLSTATE();
569	return (0);
570	}
571	break;
572	}
573	}
574
575	if (p != NULL) {
576	/*
577	* If p != NULL, we want to search the parentage tree
578	* for a matching OpenOwner and use that.
579	*/
580	if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
581	nfscl_filllockowner(NULL, own, F_POSIX);
582	else
583	nfscl_filllockowner(p->td_proc, own, F_POSIX);
584	lp = NULL;
585	error = nfscl_getopen(&clp->nfsc_owner, nfhp, fhlen, own, own,
586	mode, &lp, &op);
587	if (error == 0 && lp != NULL && fords == 0) {
588	/* Don't return a lock stateid for a DS. */
589	stateidp->seqid =
590	lp->nfsl_stateid.seqid;
591	stateidp->other[0] =
592	lp->nfsl_stateid.other[0];
593	stateidp->other[1] =
594	lp->nfsl_stateid.other[1];
595	stateidp->other[2] =
596	lp->nfsl_stateid.other[2];
597	NFSUNLOCKCLSTATE();
598	return (0);
599	}
600	}
601	if (op == NULL) {
602	/* If not found, just look for any OpenOwner that will work. */
603	top = NULL;
604	done = 0;
605	owp = LIST_FIRST(&clp->nfsc_owner);
606	while (!done && owp != NULL) {
607	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
608	if (op->nfso_fhlen == fhlen &&
609	!NFSBCMP(op->nfso_fh, nfhp, fhlen)) {
610	if (top == NULL && (op->nfso_mode &
611	NFSV4OPEN_ACCESSWRITE) != 0 &&
612	(mode & NFSV4OPEN_ACCESSREAD) != 0)
613	top = op;
614	if ((mode & op->nfso_mode) == mode) {
615	done = 1;
616	break;
617	}
618	}
619	}
620	if (!done)
621	owp = LIST_NEXT(owp, nfsow_list);
622	}
623	if (!done) {
624	NFSCL_DEBUG(2, "openmode top=%p\n", top);
625	if (top == NULL \|\| NFSHASOPENMODE(nmp)) {
626	NFSUNLOCKCLSTATE();
627	return (ENOENT);
628	} else
629	op = top;
630	}
631	/*
632	* For read aheads or write behinds, use the open cred.
633	* A read ahead or write behind is indicated by p == NULL.
634	*/
635	if (p == NULL)
636	newnfs_copycred(&op->nfso_cred, cred);
637	}
638
639	/*
640	* No lock stateid, so return the open stateid.
641	*/
642	stateidp->seqid = op->nfso_stateid.seqid;
643	stateidp->other[0] = op->nfso_stateid.other[0];
644	stateidp->other[1] = op->nfso_stateid.other[1];
645	stateidp->other[2] = op->nfso_stateid.other[2];
646	NFSUNLOCKCLSTATE();
647	return (0);
648	}
649
650	/*
651	* Search for a matching file, mode and, optionally, lockowner.
652	*/
653	static int
654	nfscl_getopen(struct nfsclownerhead ohp, u_int8_t nfhp, int fhlen,
655	u_int8_t openown, u_int8_t lockown, u_int32_t mode,
656	struct nfscllockowner lpp, struct nfsclopen opp)
657	{
658	struct nfsclowner *owp;
659	struct nfsclopen op, rop, *rop2;
660	struct nfscllockowner *lp;
661	int keep_looping;
662
663	if (lpp != NULL)
664	*lpp = NULL;
665	/*
666	* rop will be set to the open to be returned. There are three
667	* variants of this, all for an open of the correct file:
668	* 1 - A match of lockown.
669	* 2 - A match of the openown, when no lockown match exists.
670	* 3 - A match for any open, if no openown or lockown match exists.
671	* Looking for #2 over #3 probably isn't necessary, but since
672	* RFC3530 is vague w.r.t. the relationship between openowners and
673	* lockowners, I think this is the safer way to go.
674	*/
675	rop = NULL;
676	rop2 = NULL;
677	keep_looping = 1;
678	/* Search the client list */
679	owp = LIST_FIRST(ohp);
680	while (owp != NULL && keep_looping != 0) {
681	/* and look for the correct open */
682	op = LIST_FIRST(&owp->nfsow_open);
683	while (op != NULL && keep_looping != 0) {
684	if (op->nfso_fhlen == fhlen &&
685	!NFSBCMP(op->nfso_fh, nfhp, fhlen)
686	&& (op->nfso_mode & mode) == mode) {
687	if (lpp != NULL) {
688	/* Now look for a matching lockowner. */
689	LIST_FOREACH(lp, &op->nfso_lock,
690	nfsl_list) {
691	if (!NFSBCMP(lp->nfsl_owner,
692	lockown,
693	NFSV4CL_LOCKNAMELEN)) {
694	*lpp = lp;
695	rop = op;
696	keep_looping = 0;
697	break;
698	}
699	}
700	}
701	if (rop == NULL && !NFSBCMP(owp->nfsow_owner,
702	openown, NFSV4CL_LOCKNAMELEN)) {
703	rop = op;
704	if (lpp == NULL)
705	keep_looping = 0;
706	}
707	if (rop2 == NULL)
708	rop2 = op;
709	}
710	op = LIST_NEXT(op, nfso_list);
711	}
712	owp = LIST_NEXT(owp, nfsow_list);
713	}
714	if (rop == NULL)
715	rop = rop2;
716	if (rop == NULL)
717	return (EBADF);
718	*opp = rop;
719	return (0);
720	}
721
722	/*
723	* Release use of an open owner. Called when open operations are done
724	* with the open owner.
725	*/
726	APPLESTATIC void
727	nfscl_ownerrelease(struct nfsmount nmp, struct nfsclowner owp,
728	__unused int error, __unused int candelete, int unlocked)
729	{
730
731	if (owp == NULL)
732	return;
733	NFSLOCKCLSTATE();
734	if (unlocked == 0) {
735	if (NFSHASONEOPENOWN(nmp))
736	nfsv4_relref(&owp->nfsow_rwlock);
737	else
738	nfscl_lockunlock(&owp->nfsow_rwlock);
739	}
740	nfscl_clrelease(owp->nfsow_clp);
741	NFSUNLOCKCLSTATE();
742	}
743
744	/*
745	* Release use of an open structure under an open owner.
746	*/
747	APPLESTATIC void
748	nfscl_openrelease(struct nfsmount nmp, struct nfsclopen op, int error,
749	int candelete)
750	{
751	struct nfsclclient *clp;
752	struct nfsclowner *owp;
753
754	if (op == NULL)
755	return;
756	NFSLOCKCLSTATE();
757	owp = op->nfso_own;
758	if (NFSHASONEOPENOWN(nmp))
759	nfsv4_relref(&owp->nfsow_rwlock);
760	else
761	nfscl_lockunlock(&owp->nfsow_rwlock);
762	clp = owp->nfsow_clp;
763	if (error && candelete && op->nfso_opencnt == 0)
764	nfscl_freeopen(op, 0);
765	nfscl_clrelease(clp);
766	NFSUNLOCKCLSTATE();
767	}
768
769	/*
770	* Called to get a clientid structure. It will optionally lock the
771	* client data structures to do the SetClientId/SetClientId_confirm,
772	* but will release that lock and return the clientid with a reference
773	* count on it.
774	* If the "cred" argument is NULL, a new clientid should not be created.
775	* If the "p" argument is NULL, a SetClientID/SetClientIDConfirm cannot
776	* be done.
777	* The start_renewthread argument tells nfscl_getcl() to start a renew
778	* thread if this creates a new clp.
779	* It always clpp with a reference count on it, unless returning an error.
780	*/
781	APPLESTATIC int
782	nfscl_getcl(struct mount mp, struct ucred cred, NFSPROC_T *p,
783	int start_renewthread, struct nfsclclient **clpp)
784	{
785	struct nfsclclient *clp;
786	struct nfsclclient *newclp = NULL;
787	struct nfsmount *nmp;
788	char uuid[HOSTUUIDLEN];
789	int igotlock = 0, error, trystalecnt, clidinusedelay, i;
790	u_int16_t idlen = 0;
791
792	nmp = VFSTONFS(mp);
793	if (cred != NULL) {
794	getcredhostuuid(cred, uuid, sizeof uuid);
795	idlen = strlen(uuid);
796	if (idlen > 0)
797	idlen += sizeof (u_int64_t);
798	else
799	idlen += sizeof (u_int64_t) + 16; /* 16 random bytes */
800	newclp = malloc(
801	sizeof (struct nfsclclient) + idlen - 1, M_NFSCLCLIENT,
802	M_WAITOK \| M_ZERO);
803	}
804	NFSLOCKCLSTATE();
805	/*
806	* If a forced dismount is already in progress, don't
807	* allocate a new clientid and get out now. For the case where
808	* clp != NULL, this is a harmless optimization.
809	*/
810	if (NFSCL_FORCEDISM(mp)) {
811	NFSUNLOCKCLSTATE();
812	if (newclp != NULL)
813	free(newclp, M_NFSCLCLIENT);
814	return (EBADF);
815	}
816	clp = nmp->nm_clp;
817	if (clp == NULL) {
818	if (newclp == NULL) {
819	NFSUNLOCKCLSTATE();
820	return (EACCES);
821	}
822	clp = newclp;
823	clp->nfsc_idlen = idlen;
824	LIST_INIT(&clp->nfsc_owner);
825	TAILQ_INIT(&clp->nfsc_deleg);
826	TAILQ_INIT(&clp->nfsc_layout);
827	LIST_INIT(&clp->nfsc_devinfo);
828	for (i = 0; i < NFSCLDELEGHASHSIZE; i++)
829	LIST_INIT(&clp->nfsc_deleghash[i]);
830	for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
831	LIST_INIT(&clp->nfsc_layouthash[i]);
832	clp->nfsc_flags = NFSCLFLAGS_INITED;
833	clp->nfsc_clientidrev = 1;
834	clp->nfsc_cbident = nfscl_nextcbident();
835	nfscl_fillclid(nmp->nm_clval, uuid, clp->nfsc_id,
836	clp->nfsc_idlen);
837	LIST_INSERT_HEAD(&nfsclhead, clp, nfsc_list);
838	nmp->nm_clp = clp;
839	clp->nfsc_nmp = nmp;
840	NFSUNLOCKCLSTATE();
841	if (start_renewthread != 0)
842	nfscl_start_renewthread(clp);
843	} else {
844	NFSUNLOCKCLSTATE();
845	if (newclp != NULL)
846	free(newclp, M_NFSCLCLIENT);
847	}
848	NFSLOCKCLSTATE();
849	while ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0 && !igotlock &&
850	!NFSCL_FORCEDISM(mp))
851	igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
852	NFSCLSTATEMUTEXPTR, mp);
853	if (igotlock == 0) {
854	/*
855	* Call nfsv4_lock() with "iwantlock == 0" so that it will
856	* wait for a pending exclusive lock request. This gives the
857	* exclusive lock request priority over this shared lock
858	* request.
859	* An exclusive lock on nfsc_lock is used mainly for server
860	* crash recoveries.
861	*/
862	nfsv4_lock(&clp->nfsc_lock, 0, NULL, NFSCLSTATEMUTEXPTR, mp);
863	nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
864	}
865	if (igotlock == 0 && NFSCL_FORCEDISM(mp)) {
866	/*
867	* Both nfsv4_lock() and nfsv4_getref() know to check
868	* for NFSCL_FORCEDISM() and return without sleeping to
869	* wait for the exclusive lock to be released, since it
870	* might be held by nfscl_umount() and we need to get out
871	* now for that case and not wait until nfscl_umount()
872	* releases it.
873	*/
874	NFSUNLOCKCLSTATE();
875	return (EBADF);
876	}
877	NFSUNLOCKCLSTATE();
878
879	/*
880	* If it needs a clientid, do the setclientid now.
881	*/
882	if ((clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID) == 0) {
883	if (!igotlock)
884	panic("nfscl_clget");
885	if (p == NULL \|\| cred == NULL) {
886	NFSLOCKCLSTATE();
887	nfsv4_unlock(&clp->nfsc_lock, 0);
888	NFSUNLOCKCLSTATE();
889	return (EACCES);
890	}
891	/*
892	* If RFC3530 Sec. 14.2.33 is taken literally,
893	* NFSERR_CLIDINUSE will be returned persistently for the
894	* case where a new mount of the same file system is using
895	* a different principal. In practice, NFSERR_CLIDINUSE is
896	* only returned when there is outstanding unexpired state
897	* on the clientid. As such, try for twice the lease
898	* interval, if we know what that is. Otherwise, make a
899	* wild ass guess.
900	* The case of returning NFSERR_STALECLIENTID is far less
901	* likely, but might occur if there is a significant delay
902	* between doing the SetClientID and SetClientIDConfirm Ops,
903	* such that the server throws away the clientid before
904	* receiving the SetClientIDConfirm.
905	*/
906	if (clp->nfsc_renew > 0)
907	clidinusedelay = NFSCL_LEASE(clp->nfsc_renew) * 2;
908	else
909	clidinusedelay = 120;
910	trystalecnt = 3;
911	do {
912	error = nfsrpc_setclient(nmp, clp, 0, cred, p);
913	if (error == NFSERR_STALECLIENTID \|\|
914	error == NFSERR_STALEDONTRECOVER \|\|
915	error == NFSERR_BADSESSION \|\|
916	error == NFSERR_CLIDINUSE) {
917	(void) nfs_catnap(PZERO, error, "nfs_setcl");
918	}
919	} while (((error == NFSERR_STALECLIENTID \|\|
920	error == NFSERR_BADSESSION \|\|
921	error == NFSERR_STALEDONTRECOVER) && --trystalecnt > 0) \|\|
922	(error == NFSERR_CLIDINUSE && --clidinusedelay > 0));
923	if (error) {
924	NFSLOCKCLSTATE();
925	nfsv4_unlock(&clp->nfsc_lock, 0);
926	NFSUNLOCKCLSTATE();
927	return (error);
928	}
929	clp->nfsc_flags \|= NFSCLFLAGS_HASCLIENTID;
930	}
931	if (igotlock) {
932	NFSLOCKCLSTATE();
933	nfsv4_unlock(&clp->nfsc_lock, 1);
934	NFSUNLOCKCLSTATE();
935	}
936
937	*clpp = clp;
938	return (0);
939	}
940
941	/*
942	* Get a reference to a clientid and return it, if valid.
943	*/
944	APPLESTATIC struct nfsclclient *
945	nfscl_findcl(struct nfsmount *nmp)
946	{
947	struct nfsclclient *clp;
948
949	clp = nmp->nm_clp;
950	if (clp == NULL \|\| !(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID))
951	return (NULL);
952	return (clp);
953	}
954
955	/*
956	* Release the clientid structure. It may be locked or reference counted.
957	*/
958	static void
959	nfscl_clrelease(struct nfsclclient *clp)
960	{
961
962	if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
963	nfsv4_unlock(&clp->nfsc_lock, 0);
964	else
965	nfsv4_relref(&clp->nfsc_lock);
966	}
967
968	/*
969	* External call for nfscl_clrelease.
970	*/
971	APPLESTATIC void
972	nfscl_clientrelease(struct nfsclclient *clp)
973	{
974
975	NFSLOCKCLSTATE();
976	if (clp->nfsc_lock.nfslock_lock & NFSV4LOCK_LOCK)
977	nfsv4_unlock(&clp->nfsc_lock, 0);
978	else
979	nfsv4_relref(&clp->nfsc_lock);
980	NFSUNLOCKCLSTATE();
981	}
982
983	/*
984	* Called when wanting to lock a byte region.
985	*/
986	APPLESTATIC int
987	nfscl_getbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
988	short type, struct ucred cred, NFSPROC_T p, struct nfsclclient *rclp,
989	int recovery, void id, int flags, u_int8_t rownp, u_int8_t *ropenownp,
990	struct nfscllockowner *lpp, int newonep, int *donelocallyp)
991	{
992	struct nfscllockowner *lp;
993	struct nfsclopen *op;
994	struct nfsclclient *clp;
995	struct nfscllockowner *nlp;
996	struct nfscllock nlop, otherlop;
997	struct nfscldeleg dp = NULL, ldp = NULL;
998	struct nfscllockownerhead *lhp = NULL;
999	struct nfsnode *np;
1000	u_int8_t own[NFSV4CL_LOCKNAMELEN], *ownp, openown[NFSV4CL_LOCKNAMELEN];
1001	u_int8_t *openownp;
1002	int error = 0, ret, donelocally = 0;
1003	u_int32_t mode;
1004
1005	/* For Lock Ops, the open mode doesn't matter, so use 0 to match any. */
1006	mode = 0;
1007	np = VTONFS(vp);
1008	*lpp = NULL;
1009	lp = NULL;
1010	*newonep = 0;
1011	*donelocallyp = 0;
1012
1013	/*
1014	* Might need these, so MALLOC them now, to
1015	* avoid a tsleep() in MALLOC later.
1016	*/
1017	nlp = malloc(
1018	sizeof (struct nfscllockowner), M_NFSCLLOCKOWNER, M_WAITOK);
1019	otherlop = malloc(
1020	sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1021	nlop = malloc(
1022	sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1023	nlop->nfslo_type = type;
1024	nlop->nfslo_first = off;
1025	if (len == NFS64BITSSET) {
1026	nlop->nfslo_end = NFS64BITSSET;
1027	} else {
1028	nlop->nfslo_end = off + len;
1029	if (nlop->nfslo_end <= nlop->nfslo_first)
1030	error = NFSERR_INVAL;
1031	}
1032
1033	if (!error) {
1034	if (recovery)
1035	clp = rclp;
1036	else
1037	error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
1038	}
1039	if (error) {
1040	free(nlp, M_NFSCLLOCKOWNER);
1041	free(otherlop, M_NFSCLLOCK);
1042	free(nlop, M_NFSCLLOCK);
1043	return (error);
1044	}
1045
1046	op = NULL;
1047	if (recovery) {
1048	ownp = rownp;
1049	openownp = ropenownp;
1050	} else {
1051	nfscl_filllockowner(id, own, flags);
1052	ownp = own;
1053	if (NFSHASONEOPENOWN(VFSTONFS(vnode_mount(vp))))
1054	nfscl_filllockowner(NULL, openown, F_POSIX);
1055	else
1056	nfscl_filllockowner(p->td_proc, openown, F_POSIX);
1057	openownp = openown;
1058	}
1059	if (!recovery) {
1060	NFSLOCKCLSTATE();
1061	/*
1062	* First, search for a delegation. If one exists for this file,
1063	* the lock can be done locally against it, so long as there
1064	* isn't a local lock conflict.
1065	*/
1066	ldp = dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1067	np->n_fhp->nfh_len);
1068	/* Just sanity check for correct type of delegation */
1069	if (dp != NULL && ((dp->nfsdl_flags &
1070	(NFSCLDL_RECALL \| NFSCLDL_DELEGRET)) != 0 \|\|
1071	(type == F_WRLCK &&
1072	(dp->nfsdl_flags & NFSCLDL_WRITE) == 0)))
1073	dp = NULL;
1074	}
1075	if (dp != NULL) {
1076	/* Now, find an open and maybe a lockowner. */
1077	ret = nfscl_getopen(&dp->nfsdl_owner, np->n_fhp->nfh_fh,
1078	np->n_fhp->nfh_len, openownp, ownp, mode, NULL, &op);
1079	if (ret)
1080	ret = nfscl_getopen(&clp->nfsc_owner,
1081	np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1082	ownp, mode, NULL, &op);
1083	if (!ret) {
1084	lhp = &dp->nfsdl_lock;
1085	TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
1086	TAILQ_INSERT_HEAD(&clp->nfsc_deleg, dp, nfsdl_list);
1087	dp->nfsdl_timestamp = NFSD_MONOSEC + 120;
1088	donelocally = 1;
1089	} else {
1090	dp = NULL;
1091	}
1092	}
1093	if (!donelocally) {
1094	/*
1095	* Get the related Open and maybe lockowner.
1096	*/
1097	error = nfscl_getopen(&clp->nfsc_owner,
1098	np->n_fhp->nfh_fh, np->n_fhp->nfh_len, openownp,
1099	ownp, mode, &lp, &op);
1100	if (!error)
1101	lhp = &op->nfso_lock;
1102	}
1103	if (!error && !recovery)
1104	error = nfscl_localconflict(clp, np->n_fhp->nfh_fh,
1105	np->n_fhp->nfh_len, nlop, ownp, ldp, NULL);
1106	if (error) {
1107	if (!recovery) {
1108	nfscl_clrelease(clp);
1109	NFSUNLOCKCLSTATE();
1110	}
1111	free(nlp, M_NFSCLLOCKOWNER);
1112	free(otherlop, M_NFSCLLOCK);
1113	free(nlop, M_NFSCLLOCK);
1114	return (error);
1115	}
1116
1117	/*
1118	* Ok, see if a lockowner exists and create one, as required.
1119	*/
1120	if (lp == NULL)
1121	LIST_FOREACH(lp, lhp, nfsl_list) {
1122	if (!NFSBCMP(lp->nfsl_owner, ownp, NFSV4CL_LOCKNAMELEN))
1123	break;
1124	}
1125	if (lp == NULL) {
1126	NFSBCOPY(ownp, nlp->nfsl_owner, NFSV4CL_LOCKNAMELEN);
1127	if (recovery)
1128	NFSBCOPY(ropenownp, nlp->nfsl_openowner,
1129	NFSV4CL_LOCKNAMELEN);
1130	else
1131	NFSBCOPY(op->nfso_own->nfsow_owner, nlp->nfsl_openowner,
1132	NFSV4CL_LOCKNAMELEN);
1133	nlp->nfsl_seqid = 0;
1134	nlp->nfsl_lockflags = flags;
1135	nlp->nfsl_inprog = NULL;
1136	nfscl_lockinit(&nlp->nfsl_rwlock);
1137	LIST_INIT(&nlp->nfsl_lock);
1138	if (donelocally) {
1139	nlp->nfsl_open = NULL;
1140	nfsstatsv1.cllocallockowners++;
1141	} else {
1142	nlp->nfsl_open = op;
1143	nfsstatsv1.cllockowners++;
1144	}
1145	LIST_INSERT_HEAD(lhp, nlp, nfsl_list);
1146	lp = nlp;
1147	nlp = NULL;
1148	*newonep = 1;
1149	}
1150
1151	/*
1152	* Now, update the byte ranges for locks.
1153	*/
1154	ret = nfscl_updatelock(lp, &nlop, &otherlop, donelocally);
1155	if (!ret)
1156	donelocally = 1;
1157	if (donelocally) {
1158	*donelocallyp = 1;
1159	if (!recovery)
1160	nfscl_clrelease(clp);
1161	} else {
1162	/*
1163	* Serial modifications on the lock owner for multiple threads
1164	* for the same process using a read/write lock.
1165	*/
1166	if (!recovery)
1167	nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1168	}
1169	if (!recovery)
1170	NFSUNLOCKCLSTATE();
1171
1172	if (nlp)
1173	free(nlp, M_NFSCLLOCKOWNER);
1174	if (nlop)
1175	free(nlop, M_NFSCLLOCK);
1176	if (otherlop)
1177	free(otherlop, M_NFSCLLOCK);
1178
1179	*lpp = lp;
1180	return (0);
1181	}
1182
1183	/*
1184	* Called to unlock a byte range, for LockU.
1185	*/
1186	APPLESTATIC int
1187	nfscl_relbytelock(vnode_t vp, u_int64_t off, u_int64_t len,
1188	__unused struct ucred cred, NFSPROC_T p, int callcnt,
1189	struct nfsclclient clp, void id, int flags,
1190	struct nfscllockowner *lpp, int dorpcp)
1191	{
1192	struct nfscllockowner *lp;
1193	struct nfsclowner *owp;
1194	struct nfsclopen *op;
1195	struct nfscllock nlop, other_lop = NULL;
1196	struct nfscldeleg *dp;
1197	struct nfsnode *np;
1198	u_int8_t own[NFSV4CL_LOCKNAMELEN];
1199	int ret = 0, fnd;
1200
1201	np = VTONFS(vp);
1202	*lpp = NULL;
1203	*dorpcp = 0;
1204
1205	/*
1206	* Might need these, so MALLOC them now, to
1207	* avoid a tsleep() in MALLOC later.
1208	*/
1209	nlop = malloc(
1210	sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1211	nlop->nfslo_type = F_UNLCK;
1212	nlop->nfslo_first = off;
1213	if (len == NFS64BITSSET) {
1214	nlop->nfslo_end = NFS64BITSSET;
1215	} else {
1216	nlop->nfslo_end = off + len;
1217	if (nlop->nfslo_end <= nlop->nfslo_first) {
1218	free(nlop, M_NFSCLLOCK);
1219	return (NFSERR_INVAL);
1220	}
1221	}
1222	if (callcnt == 0) {
1223	other_lop = malloc(
1224	sizeof (struct nfscllock), M_NFSCLLOCK, M_WAITOK);
1225	other_lop = nlop;
1226	}
1227	nfscl_filllockowner(id, own, flags);
1228	dp = NULL;
1229	NFSLOCKCLSTATE();
1230	if (callcnt == 0)
1231	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
1232	np->n_fhp->nfh_len);
1233
1234	/*
1235	* First, unlock any local regions on a delegation.
1236	*/
1237	if (dp != NULL) {
1238	/* Look for this lockowner. */
1239	LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1240	if (!NFSBCMP(lp->nfsl_owner, own,
1241	NFSV4CL_LOCKNAMELEN))
1242	break;
1243	}
1244	if (lp != NULL)
1245	/* Use other_lop, so nlop is still available */
1246	(void)nfscl_updatelock(lp, &other_lop, NULL, 1);
1247	}
1248
1249	/*
1250	* Now, find a matching open/lockowner that hasn't already been done,
1251	* as marked by nfsl_inprog.
1252	*/
1253	lp = NULL;
1254	fnd = 0;
1255	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1256	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1257	if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1258	!NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1259	LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1260	if (lp->nfsl_inprog == NULL &&
1261	!NFSBCMP(lp->nfsl_owner, own,
1262	NFSV4CL_LOCKNAMELEN)) {
1263	fnd = 1;
1264	break;
1265	}
1266	}
1267	if (fnd)
1268	break;
1269	}
1270	}
1271	if (fnd)
1272	break;
1273	}
1274
1275	if (lp != NULL) {
1276	ret = nfscl_updatelock(lp, &nlop, NULL, 0);
1277	if (ret)
1278	*dorpcp = 1;
1279	/*
1280	* Serial modifications on the lock owner for multiple
1281	* threads for the same process using a read/write lock.
1282	*/
1283	lp->nfsl_inprog = p;
1284	nfscl_lockexcl(&lp->nfsl_rwlock, NFSCLSTATEMUTEXPTR);
1285	*lpp = lp;
1286	}
1287	NFSUNLOCKCLSTATE();
1288	if (nlop)
1289	free(nlop, M_NFSCLLOCK);
1290	if (other_lop)
1291	free(other_lop, M_NFSCLLOCK);
1292	return (0);
1293	}
1294
1295	/*
1296	* Release all lockowners marked in progess for this process and file.
1297	*/
1298	APPLESTATIC void
1299	nfscl_releasealllocks(struct nfsclclient clp, vnode_t vp, NFSPROC_T p,
1300	void *id, int flags)
1301	{
1302	struct nfsclowner *owp;
1303	struct nfsclopen *op;
1304	struct nfscllockowner *lp;
1305	struct nfsnode *np;
1306	u_int8_t own[NFSV4CL_LOCKNAMELEN];
1307
1308	np = VTONFS(vp);
1309	nfscl_filllockowner(id, own, flags);
1310	NFSLOCKCLSTATE();
1311	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1312	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1313	if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1314	!NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1315	LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1316	if (lp->nfsl_inprog == p &&
1317	!NFSBCMP(lp->nfsl_owner, own,
1318	NFSV4CL_LOCKNAMELEN)) {
1319	lp->nfsl_inprog = NULL;
1320	nfscl_lockunlock(&lp->nfsl_rwlock);
1321	}
1322	}
1323	}
1324	}
1325	}
1326	nfscl_clrelease(clp);
1327	NFSUNLOCKCLSTATE();
1328	}
1329
1330	/*
1331	* Called to find out if any bytes within the byte range specified are
1332	* write locked by the calling process. Used to determine if flushing
1333	* is required before a LockU.
1334	* If in doubt, return 1, so the flush will occur.
1335	*/
1336	APPLESTATIC int
1337	nfscl_checkwritelocked(vnode_t vp, struct flock *fl,
1338	struct ucred cred, NFSPROC_T p, void *id, int flags)
1339	{
1340	struct nfsclowner *owp;
1341	struct nfscllockowner *lp;
1342	struct nfsclopen *op;
1343	struct nfsclclient *clp;
1344	struct nfscllock *lop;
1345	struct nfscldeleg *dp;
1346	struct nfsnode *np;
1347	u_int64_t off, end;
1348	u_int8_t own[NFSV4CL_LOCKNAMELEN];
1349	int error = 0;
1350
1351	np = VTONFS(vp);
1352	switch (fl->l_whence) {
1353	case SEEK_SET:
1354	case SEEK_CUR:
1355	/*
1356	* Caller is responsible for adding any necessary offset
1357	* when SEEK_CUR is used.
1358	*/
1359	off = fl->l_start;
1360	break;
1361	case SEEK_END:
1362	off = np->n_size + fl->l_start;
1363	break;
1364	default:
1365	return (1);
1366	}
1367	if (fl->l_len != 0) {
1368	end = off + fl->l_len;
1369	if (end < off)
1370	return (1);
1371	} else {
1372	end = NFS64BITSSET;
1373	}
1374
1375	error = nfscl_getcl(vnode_mount(vp), cred, p, 1, &clp);
1376	if (error)
1377	return (1);
1378	nfscl_filllockowner(id, own, flags);
1379	NFSLOCKCLSTATE();
1380
1381	/*
1382	* First check the delegation locks.
1383	*/
1384	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
1385	if (dp != NULL) {
1386	LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
1387	if (!NFSBCMP(lp->nfsl_owner, own,
1388	NFSV4CL_LOCKNAMELEN))
1389	break;
1390	}
1391	if (lp != NULL) {
1392	LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1393	if (lop->nfslo_first >= end)
1394	break;
1395	if (lop->nfslo_end <= off)
1396	continue;
1397	if (lop->nfslo_type == F_WRLCK) {
1398	nfscl_clrelease(clp);
1399	NFSUNLOCKCLSTATE();
1400	return (1);
1401	}
1402	}
1403	}
1404	}
1405
1406	/*
1407	* Now, check state against the server.
1408	*/
1409	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
1410	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1411	if (op->nfso_fhlen == np->n_fhp->nfh_len &&
1412	!NFSBCMP(op->nfso_fh, np->n_fhp->nfh_fh, op->nfso_fhlen)) {
1413	LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1414	if (!NFSBCMP(lp->nfsl_owner, own,
1415	NFSV4CL_LOCKNAMELEN))
1416	break;
1417	}
1418	if (lp != NULL) {
1419	LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
1420	if (lop->nfslo_first >= end)
1421	break;
1422	if (lop->nfslo_end <= off)
1423	continue;
1424	if (lop->nfslo_type == F_WRLCK) {
1425	nfscl_clrelease(clp);
1426	NFSUNLOCKCLSTATE();
1427	return (1);
1428	}
1429	}
1430	}
1431	}
1432	}
1433	}
1434	nfscl_clrelease(clp);
1435	NFSUNLOCKCLSTATE();
1436	return (0);
1437	}
1438
1439	/*
1440	* Release a byte range lock owner structure.
1441	*/
1442	APPLESTATIC void
1443	nfscl_lockrelease(struct nfscllockowner *lp, int error, int candelete)
1444	{
1445	struct nfsclclient *clp;
1446
1447	if (lp == NULL)
1448	return;
1449	NFSLOCKCLSTATE();
1450	clp = lp->nfsl_open->nfso_own->nfsow_clp;
1451	if (error != 0 && candelete &&
1452	(lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED) == 0)
1453	nfscl_freelockowner(lp, 0);
1454	else
1455	nfscl_lockunlock(&lp->nfsl_rwlock);
1456	nfscl_clrelease(clp);
1457	NFSUNLOCKCLSTATE();
1458	}
1459
1460	/*
1461	* Free up an open structure and any associated byte range lock structures.
1462	*/
1463	APPLESTATIC void
1464	nfscl_freeopen(struct nfsclopen *op, int local)
1465	{
1466
1467	LIST_REMOVE(op, nfso_list);
1468	nfscl_freealllocks(&op->nfso_lock, local);
1469	free(op, M_NFSCLOPEN);
1470	if (local)
1471	nfsstatsv1.cllocalopens--;
1472	else
1473	nfsstatsv1.clopens--;
1474	}
1475
1476	/*
1477	* Free up all lock owners and associated locks.
1478	*/
1479	static void
1480	nfscl_freealllocks(struct nfscllockownerhead *lhp, int local)
1481	{
1482	struct nfscllockowner lp, nlp;
1483
1484	LIST_FOREACH_SAFE(lp, lhp, nfsl_list, nlp) {
1485	if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1486	panic("nfscllckw");
1487	nfscl_freelockowner(lp, local);
1488	}
1489	}
1490
1491	/*
1492	* Called for an Open when NFSERR_EXPIRED is received from the server.
1493	* If there are no byte range locks nor a Share Deny lost, try to do a
1494	* fresh Open. Otherwise, free the open.
1495	*/
1496	static int
1497	nfscl_expireopen(struct nfsclclient clp, struct nfsclopen op,
1498	struct nfsmount nmp, struct ucred cred, NFSPROC_T *p)
1499	{
1500	struct nfscllockowner *lp;
1501	struct nfscldeleg *dp;
1502	int mustdelete = 0, error;
1503
1504	/*
1505	* Look for any byte range lock(s).
1506	*/
1507	LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
1508	if (!LIST_EMPTY(&lp->nfsl_lock)) {
1509	mustdelete = 1;
1510	break;
1511	}
1512	}
1513
1514	/*
1515	* If no byte range lock(s) nor a Share deny, try to re-open.
1516	*/
1517	if (!mustdelete && (op->nfso_mode & NFSLCK_DENYBITS) == 0) {
1518	newnfs_copycred(&op->nfso_cred, cred);
1519	dp = NULL;
1520	error = nfsrpc_reopen(nmp, op->nfso_fh,
1521	op->nfso_fhlen, op->nfso_mode, op, &dp, cred, p);
1522	if (error) {
1523	mustdelete = 1;
1524	if (dp != NULL) {
1525	free(dp, M_NFSCLDELEG);
1526	dp = NULL;
1527	}
1528	}
1529	if (dp != NULL)
1530	nfscl_deleg(nmp->nm_mountp, clp, op->nfso_fh,
1531	op->nfso_fhlen, cred, p, &dp);
1532	}
1533
1534	/*
1535	* If a byte range lock or Share deny or couldn't re-open, free it.
1536	*/
1537	if (mustdelete)
1538	nfscl_freeopen(op, 0);
1539	return (mustdelete);
1540	}
1541
1542	/*
1543	* Free up an open owner structure.
1544	*/
1545	static void
1546	nfscl_freeopenowner(struct nfsclowner *owp, int local)
1547	{
1548
1549	LIST_REMOVE(owp, nfsow_list);
1550	free(owp, M_NFSCLOWNER);
1551	if (local)
1552	nfsstatsv1.cllocalopenowners--;
1553	else
1554	nfsstatsv1.clopenowners--;
1555	}
1556
1557	/*
1558	* Free up a byte range lock owner structure.
1559	*/
1560	APPLESTATIC void
1561	nfscl_freelockowner(struct nfscllockowner *lp, int local)
1562	{
1563	struct nfscllock lop, nlop;
1564
1565	LIST_REMOVE(lp, nfsl_list);
1566	LIST_FOREACH_SAFE(lop, &lp->nfsl_lock, nfslo_list, nlop) {
1567	nfscl_freelock(lop, local);
1568	}
1569	free(lp, M_NFSCLLOCKOWNER);
1570	if (local)
1571	nfsstatsv1.cllocallockowners--;
1572	else
1573	nfsstatsv1.cllockowners--;
1574	}
1575
1576	/*
1577	* Free up a byte range lock structure.
1578	*/
1579	APPLESTATIC void
1580	nfscl_freelock(struct nfscllock *lop, int local)
1581	{
1582
1583	LIST_REMOVE(lop, nfslo_list);
1584	free(lop, M_NFSCLLOCK);
1585	if (local)
1586	nfsstatsv1.cllocallocks--;
1587	else
1588	nfsstatsv1.cllocks--;
1589	}
1590
1591	/*
1592	* Clean out the state related to a delegation.
1593	*/
1594	static void
1595	nfscl_cleandeleg(struct nfscldeleg *dp)
1596	{
1597	struct nfsclowner owp, nowp;
1598	struct nfsclopen *op;
1599
1600	LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
1601	op = LIST_FIRST(&owp->nfsow_open);
1602	if (op != NULL) {
1603	if (LIST_NEXT(op, nfso_list) != NULL)
1604	panic("nfscleandel");
1605	nfscl_freeopen(op, 1);
1606	}
1607	nfscl_freeopenowner(owp, 1);
1608	}
1609	nfscl_freealllocks(&dp->nfsdl_lock, 1);
1610	}
1611
1612	/*
1613	* Free a delegation.
1614	*/
1615	static void
1616	nfscl_freedeleg(struct nfscldeleghead hdp, struct nfscldeleg dp)
1617	{
1618
1619	TAILQ_REMOVE(hdp, dp, nfsdl_list);
1620	LIST_REMOVE(dp, nfsdl_hash);
1621	free(dp, M_NFSCLDELEG);
1622	nfsstatsv1.cldelegates--;
1623	nfscl_delegcnt--;
1624	}
1625
1626	/*
1627	* Free up all state related to this client structure.
1628	*/
1629	static void
1630	nfscl_cleanclient(struct nfsclclient *clp)
1631	{
1632	struct nfsclowner owp, nowp;
1633	struct nfsclopen op, nop;
1634	struct nfscllayout lyp, nlyp;
1635	struct nfscldevinfo dip, ndip;
1636
1637	TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
1638	nfscl_freelayout(lyp);
1639
1640	LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip)
1641	nfscl_freedevinfo(dip);
1642
1643	/* Now, all the OpenOwners, etc. */
1644	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1645	LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1646	nfscl_freeopen(op, 0);
1647	}
1648	nfscl_freeopenowner(owp, 0);
1649	}
1650	}
1651
1652	/*
1653	* Called when an NFSERR_EXPIRED is received from the server.
1654	*/
1655	static void
1656	nfscl_expireclient(struct nfsclclient clp, struct nfsmount nmp,
1657	struct ucred cred, NFSPROC_T p)
1658	{
1659	struct nfsclowner owp, nowp, *towp;
1660	struct nfsclopen op, nop, *top;
1661	struct nfscldeleg dp, ndp;
1662	int ret, printed = 0;
1663
1664	/*
1665	* First, merge locally issued Opens into the list for the server.
1666	*/
1667	dp = TAILQ_FIRST(&clp->nfsc_deleg);
1668	while (dp != NULL) {
1669	ndp = TAILQ_NEXT(dp, nfsdl_list);
1670	owp = LIST_FIRST(&dp->nfsdl_owner);
1671	while (owp != NULL) {
1672	nowp = LIST_NEXT(owp, nfsow_list);
1673	op = LIST_FIRST(&owp->nfsow_open);
1674	if (op != NULL) {
1675	if (LIST_NEXT(op, nfso_list) != NULL)
1676	panic("nfsclexp");
1677	LIST_FOREACH(towp, &clp->nfsc_owner, nfsow_list) {
1678	if (!NFSBCMP(towp->nfsow_owner, owp->nfsow_owner,
1679	NFSV4CL_LOCKNAMELEN))
1680	break;
1681	}
1682	if (towp != NULL) {
1683	/* Merge opens in */
1684	LIST_FOREACH(top, &towp->nfsow_open, nfso_list) {
1685	if (top->nfso_fhlen == op->nfso_fhlen &&
1686	!NFSBCMP(top->nfso_fh, op->nfso_fh,
1687	op->nfso_fhlen)) {
1688	top->nfso_mode \|= op->nfso_mode;
1689	top->nfso_opencnt += op->nfso_opencnt;
1690	break;
1691	}
1692	}
1693	if (top == NULL) {
1694	/* Just add the open to the owner list */
1695	LIST_REMOVE(op, nfso_list);
1696	op->nfso_own = towp;
1697	LIST_INSERT_HEAD(&towp->nfsow_open, op, nfso_list);
1698	nfsstatsv1.cllocalopens--;
1699	nfsstatsv1.clopens++;
1700	}
1701	} else {
1702	/* Just add the openowner to the client list */
1703	LIST_REMOVE(owp, nfsow_list);
1704	owp->nfsow_clp = clp;
1705	LIST_INSERT_HEAD(&clp->nfsc_owner, owp, nfsow_list);
1706	nfsstatsv1.cllocalopenowners--;
1707	nfsstatsv1.clopenowners++;
1708	nfsstatsv1.cllocalopens--;
1709	nfsstatsv1.clopens++;
1710	}
1711	}
1712	owp = nowp;
1713	}
1714	if (!printed && !LIST_EMPTY(&dp->nfsdl_lock)) {
1715	printed = 1;
1716	printf("nfsv4 expired locks lost\n");
1717	}
1718	nfscl_cleandeleg(dp);
1719	nfscl_freedeleg(&clp->nfsc_deleg, dp);
1720	dp = ndp;
1721	}
1722	if (!TAILQ_EMPTY(&clp->nfsc_deleg))
1723	panic("nfsclexp");
1724
1725	/*
1726	* Now, try and reopen against the server.
1727	*/
1728	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1729	owp->nfsow_seqid = 0;
1730	LIST_FOREACH_SAFE(op, &owp->nfsow_open, nfso_list, nop) {
1731	ret = nfscl_expireopen(clp, op, nmp, cred, p);
1732	if (ret && !printed) {
1733	printed = 1;
1734	printf("nfsv4 expired locks lost\n");
1735	}
1736	}
1737	if (LIST_EMPTY(&owp->nfsow_open))
1738	nfscl_freeopenowner(owp, 0);
1739	}
1740	}
1741
1742	/*
1743	* This function must be called after the process represented by "own" has
1744	* exited. Must be called with CLSTATE lock held.
1745	*/
1746	static void
1747	nfscl_cleanup_common(struct nfsclclient clp, u_int8_t own)
1748	{
1749	struct nfsclowner owp, nowp;
1750	struct nfscllockowner lp, nlp;
1751	struct nfscldeleg *dp;
1752
1753	/* First, get rid of local locks on delegations. */
1754	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1755	LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1756	if (!NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
1757	if ((lp->nfsl_rwlock.nfslock_lock & NFSV4LOCK_WANTED))
1758	panic("nfscllckw");
1759	nfscl_freelockowner(lp, 1);
1760	}
1761	}
1762	}
1763	owp = LIST_FIRST(&clp->nfsc_owner);
1764	while (owp != NULL) {
1765	nowp = LIST_NEXT(owp, nfsow_list);
1766	if (!NFSBCMP(owp->nfsow_owner, own,
1767	NFSV4CL_LOCKNAMELEN)) {
1768	/*
1769	* If there are children that haven't closed the
1770	* file descriptors yet, the opens will still be
1771	* here. For that case, let the renew thread clear
1772	* out the OpenOwner later.
1773	*/
1774	if (LIST_EMPTY(&owp->nfsow_open))
1775	nfscl_freeopenowner(owp, 0);
1776	else
1777	owp->nfsow_defunct = 1;
1778	}
1779	owp = nowp;
1780	}
1781	}
1782
1783	/*
1784	* Find open/lock owners for processes that have exited.
1785	*/
1786	static void
1787	nfscl_cleanupkext(struct nfsclclient clp, struct nfscllockownerfhhead lhp)
1788	{
1789	struct nfsclowner owp, nowp;
1790	struct nfsclopen *op;
1791	struct nfscllockowner lp, nlp;
1792	struct nfscldeleg *dp;
1793
1794	NFSPROCLISTLOCK();
1795	NFSLOCKCLSTATE();
1796	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
1797	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
1798	LIST_FOREACH_SAFE(lp, &op->nfso_lock, nfsl_list, nlp) {
1799	if (LIST_EMPTY(&lp->nfsl_lock))
1800	nfscl_emptylockowner(lp, lhp);
1801	}
1802	}
1803	if (nfscl_procdoesntexist(owp->nfsow_owner))
1804	nfscl_cleanup_common(clp, owp->nfsow_owner);
1805	}
1806
1807	/*
1808	* For the single open_owner case, these lock owners need to be
1809	* checked to see if they still exist separately.
1810	* This is because nfscl_procdoesntexist() never returns true for
1811	* the single open_owner so that the above doesn't ever call
1812	* nfscl_cleanup_common().
1813	*/
1814	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
1815	LIST_FOREACH_SAFE(lp, &dp->nfsdl_lock, nfsl_list, nlp) {
1816	if (nfscl_procdoesntexist(lp->nfsl_owner))
1817	nfscl_cleanup_common(clp, lp->nfsl_owner);
1818	}
1819	}
1820	NFSUNLOCKCLSTATE();
1821	NFSPROCLISTUNLOCK();
1822	}
1823
1824	/*
1825	* Take the empty lock owner and move it to the local lhp list if the
1826	* associated process no longer exists.
1827	*/
1828	static void
1829	nfscl_emptylockowner(struct nfscllockowner *lp,
1830	struct nfscllockownerfhhead *lhp)
1831	{
1832	struct nfscllockownerfh lfhp, mylfhp;
1833	struct nfscllockowner *nlp;
1834	int fnd_it;
1835
1836	/* If not a Posix lock owner, just return. */
1837	if ((lp->nfsl_lockflags & F_POSIX) == 0)
1838	return;
1839
1840	fnd_it = 0;
1841	mylfhp = NULL;
1842	/*
1843	* First, search to see if this lock owner is already in the list.
1844	* If it is, then the associated process no longer exists.
1845	*/
1846	SLIST_FOREACH(lfhp, lhp, nfslfh_list) {
1847	if (lfhp->nfslfh_len == lp->nfsl_open->nfso_fhlen &&
1848	!NFSBCMP(lfhp->nfslfh_fh, lp->nfsl_open->nfso_fh,
1849	lfhp->nfslfh_len))
1850	mylfhp = lfhp;
1851	LIST_FOREACH(nlp, &lfhp->nfslfh_lock, nfsl_list)
1852	if (!NFSBCMP(nlp->nfsl_owner, lp->nfsl_owner,
1853	NFSV4CL_LOCKNAMELEN))
1854	fnd_it = 1;
1855	}
1856	/* If not found, check if process still exists. */
1857	if (fnd_it == 0 && nfscl_procdoesntexist(lp->nfsl_owner) == 0)
1858	return;
1859
1860	/* Move the lock owner over to the local list. */
1861	if (mylfhp == NULL) {
1862	mylfhp = malloc(sizeof(struct nfscllockownerfh), M_TEMP,
1863	M_NOWAIT);
1864	if (mylfhp == NULL)
1865	return;
1866	mylfhp->nfslfh_len = lp->nfsl_open->nfso_fhlen;
1867	NFSBCOPY(lp->nfsl_open->nfso_fh, mylfhp->nfslfh_fh,
1868	mylfhp->nfslfh_len);
1869	LIST_INIT(&mylfhp->nfslfh_lock);
1870	SLIST_INSERT_HEAD(lhp, mylfhp, nfslfh_list);
1871	}
1872	LIST_REMOVE(lp, nfsl_list);
1873	LIST_INSERT_HEAD(&mylfhp->nfslfh_lock, lp, nfsl_list);
1874	}
1875
1876	static int fake_global; /* Used to force visibility of MNTK_UNMOUNTF */
1877	/*
1878	* Called from nfs umount to free up the clientid.
1879	*/
1880	APPLESTATIC void
1881	nfscl_umount(struct nfsmount nmp, NFSPROC_T p)
1882	{
1883	struct nfsclclient *clp;
1884	struct ucred *cred;
1885	int igotlock;
1886
1887	/*
1888	* For the case that matters, this is the thread that set
1889	* MNTK_UNMOUNTF, so it will see it set. The code that follows is
1890	* done to ensure that any thread executing nfscl_getcl() after
1891	* this time, will see MNTK_UNMOUNTF set. nfscl_getcl() uses the
1892	* mutex for NFSLOCKCLSTATE(), so it is "m" for the following
1893	* explanation, courtesy of Alan Cox.
1894	* What follows is a snippet from Alan Cox's email at:
1895	* https://docs.FreeBSD.org/cgi/mid.cgi?BANLkTikR3d65zPHo9==08ZfJ2vmqZucEvw
1896	*
1897	* 1. Set MNTK_UNMOUNTF
1898	* 2. Acquire a standard FreeBSD mutex "m".
1899	* 3. Update some data structures.
1900	* 4. Release mutex "m".
1901	*
1902	* Then, other threads that acquire "m" after step 4 has occurred will
1903	* see MNTK_UNMOUNTF as set. But, other threads that beat thread X to
1904	* step 2 may or may not see MNTK_UNMOUNTF as set.
1905	*/
1906	NFSLOCKCLSTATE();
1907	if ((nmp->nm_mountp->mnt_kern_flag & MNTK_UNMOUNTF) != 0) {
1908	fake_global++;
1909	NFSUNLOCKCLSTATE();
1910	NFSLOCKCLSTATE();
1911	}
1912
1913	clp = nmp->nm_clp;
1914	if (clp != NULL) {
1915	if ((clp->nfsc_flags & NFSCLFLAGS_INITED) == 0)
1916	panic("nfscl umount");
1917
1918	/*
1919	* First, handshake with the nfscl renew thread, to terminate
1920	* it.
1921	*/
1922	clp->nfsc_flags \|= NFSCLFLAGS_UMOUNT;
1923	while (clp->nfsc_flags & NFSCLFLAGS_HASTHREAD)
1924	(void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT,
1925	"nfsclumnt", hz);
1926
1927	/*
1928	* Now, get the exclusive lock on the client state, so
1929	* that no uses of the state are still in progress.
1930	*/
1931	do {
1932	igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1933	NFSCLSTATEMUTEXPTR, NULL);
1934	} while (!igotlock);
1935	NFSUNLOCKCLSTATE();
1936
1937	/*
1938	* Free up all the state. It will expire on the server, but
1939	* maybe we should do a SetClientId/SetClientIdConfirm so
1940	* the server throws it away?
1941	*/
1942	LIST_REMOVE(clp, nfsc_list);
1943	nfscl_delegreturnall(clp, p);
1944	cred = newnfs_getcred();
1945	if (NFSHASNFSV4N(nmp)) {
1946	(void)nfsrpc_destroysession(nmp, clp, cred, p);
1947	(void)nfsrpc_destroyclient(nmp, clp, cred, p);
1948	} else
1949	(void)nfsrpc_setclient(nmp, clp, 0, cred, p);
1950	nfscl_cleanclient(clp);
1951	nmp->nm_clp = NULL;
1952	NFSFREECRED(cred);
1953	free(clp, M_NFSCLCLIENT);
1954	} else
1955	NFSUNLOCKCLSTATE();
1956	}
1957
1958	/*
1959	* This function is called when a server replies with NFSERR_STALECLIENTID
1960	* NFSERR_STALESTATEID or NFSERR_BADSESSION. It traverses the clientid lists,
1961	* doing Opens and Locks with reclaim. If these fail, it deletes the
1962	* corresponding state.
1963	*/
1964	static void
1965	nfscl_recover(struct nfsclclient clp, struct ucred cred, NFSPROC_T *p)
1966	{
1967	struct nfsclowner owp, nowp;
1968	struct nfsclopen op, nop;
1969	struct nfscllockowner lp, nlp;
1970	struct nfscllock lop, nlop;
1971	struct nfscldeleg dp, ndp, *tdp;
1972	struct nfsmount *nmp;
1973	struct ucred *tcred;
1974	struct nfsclopenhead extra_open;
1975	struct nfscldeleghead extra_deleg;
1976	struct nfsreq *rep;
1977	u_int64_t len;
1978	u_int32_t delegtype = NFSV4OPEN_DELEGATEWRITE, mode;
1979	int i, igotlock = 0, error, trycnt, firstlock;
1980	struct nfscllayout lyp, nlyp;
1981
1982	/*
1983	* First, lock the client structure, so everyone else will
1984	* block when trying to use state.
1985	*/
1986	NFSLOCKCLSTATE();
1987	clp->nfsc_flags \|= NFSCLFLAGS_RECVRINPROG;
1988	do {
1989	igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
1990	NFSCLSTATEMUTEXPTR, NULL);
1991	} while (!igotlock);
1992	NFSUNLOCKCLSTATE();
1993
1994	nmp = clp->nfsc_nmp;
1995	if (nmp == NULL)
1996	panic("nfscl recover");
1997
1998	/*
1999	* For now, just get rid of all layouts. There may be a need
2000	* to do LayoutCommit Ops with reclaim == true later.
2001	*/
2002	TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp)
2003	nfscl_freelayout(lyp);
2004	TAILQ_INIT(&clp->nfsc_layout);
2005	for (i = 0; i < NFSCLLAYOUTHASHSIZE; i++)
2006	LIST_INIT(&clp->nfsc_layouthash[i]);
2007
2008	trycnt = 5;
2009	do {
2010	error = nfsrpc_setclient(nmp, clp, 1, cred, p);
2011	} while ((error == NFSERR_STALECLIENTID \|\|
2012	error == NFSERR_BADSESSION \|\|
2013	error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2014	if (error) {
2015	NFSLOCKCLSTATE();
2016	clp->nfsc_flags &= ~(NFSCLFLAGS_RECOVER \|
2017	NFSCLFLAGS_RECVRINPROG);
2018	wakeup(&clp->nfsc_flags);
2019	nfsv4_unlock(&clp->nfsc_lock, 0);
2020	NFSUNLOCKCLSTATE();
2021	return;
2022	}
2023	clp->nfsc_flags \|= NFSCLFLAGS_HASCLIENTID;
2024	clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2025
2026	/*
2027	* Mark requests already queued on the server, so that they don't
2028	* initiate another recovery cycle. Any requests already in the
2029	* queue that handle state information will have the old stale
2030	* clientid/stateid and will get a NFSERR_STALESTATEID,
2031	* NFSERR_STALECLIENTID or NFSERR_BADSESSION reply from the server.
2032	* This will be translated to NFSERR_STALEDONTRECOVER when
2033	* R_DONTRECOVER is set.
2034	*/
2035	NFSLOCKREQ();
2036	TAILQ_FOREACH(rep, &nfsd_reqq, r_chain) {
2037	if (rep->r_nmp == nmp)
2038	rep->r_flags \|= R_DONTRECOVER;
2039	}
2040	NFSUNLOCKREQ();
2041
2042	/*
2043	* Now, mark all delegations "need reclaim".
2044	*/
2045	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list)
2046	dp->nfsdl_flags \|= NFSCLDL_NEEDRECLAIM;
2047
2048	TAILQ_INIT(&extra_deleg);
2049	LIST_INIT(&extra_open);
2050	/*
2051	* Now traverse the state lists, doing Open and Lock Reclaims.
2052	*/
2053	tcred = newnfs_getcred();
2054	owp = LIST_FIRST(&clp->nfsc_owner);
2055	while (owp != NULL) {
2056	nowp = LIST_NEXT(owp, nfsow_list);
2057	owp->nfsow_seqid = 0;
2058	op = LIST_FIRST(&owp->nfsow_open);
2059	while (op != NULL) {
2060	nop = LIST_NEXT(op, nfso_list);
2061	if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2062	/* Search for a delegation to reclaim with the open */
2063	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2064	if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2065	continue;
2066	if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2067	mode = NFSV4OPEN_ACCESSWRITE;
2068	delegtype = NFSV4OPEN_DELEGATEWRITE;
2069	} else {
2070	mode = NFSV4OPEN_ACCESSREAD;
2071	delegtype = NFSV4OPEN_DELEGATEREAD;
2072	}
2073	if ((op->nfso_mode & mode) == mode &&
2074	op->nfso_fhlen == dp->nfsdl_fhlen &&
2075	!NFSBCMP(op->nfso_fh, dp->nfsdl_fh, op->nfso_fhlen))
2076	break;
2077	}
2078	ndp = dp;
2079	if (dp == NULL)
2080	delegtype = NFSV4OPEN_DELEGATENONE;
2081	newnfs_copycred(&op->nfso_cred, tcred);
2082	error = nfscl_tryopen(nmp, NULL, op->nfso_fh,
2083	op->nfso_fhlen, op->nfso_fh, op->nfso_fhlen,
2084	op->nfso_mode, op, NULL, 0, &ndp, 1, delegtype,
2085	tcred, p);
2086	if (!error) {
2087	/* Handle any replied delegation */
2088	if (ndp != NULL && ((ndp->nfsdl_flags & NFSCLDL_WRITE)
2089	\|\| NFSMNT_RDONLY(nmp->nm_mountp))) {
2090	if ((ndp->nfsdl_flags & NFSCLDL_WRITE))
2091	mode = NFSV4OPEN_ACCESSWRITE;
2092	else
2093	mode = NFSV4OPEN_ACCESSREAD;
2094	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2095	if (!(dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM))
2096	continue;
2097	if ((op->nfso_mode & mode) == mode &&
2098	op->nfso_fhlen == dp->nfsdl_fhlen &&
2099	!NFSBCMP(op->nfso_fh, dp->nfsdl_fh,
2100	op->nfso_fhlen)) {
2101	dp->nfsdl_stateid = ndp->nfsdl_stateid;
2102	dp->nfsdl_sizelimit = ndp->nfsdl_sizelimit;
2103	dp->nfsdl_ace = ndp->nfsdl_ace;
2104	dp->nfsdl_change = ndp->nfsdl_change;
2105	dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2106	if ((ndp->nfsdl_flags & NFSCLDL_RECALL))
2107	dp->nfsdl_flags \|= NFSCLDL_RECALL;
2108	free(ndp, M_NFSCLDELEG);
2109	ndp = NULL;
2110	break;
2111	}
2112	}
2113	}
2114	if (ndp != NULL)
2115	TAILQ_INSERT_HEAD(&extra_deleg, ndp, nfsdl_list);
2116
2117	/* and reclaim all byte range locks */
2118	lp = LIST_FIRST(&op->nfso_lock);
2119	while (lp != NULL) {
2120	nlp = LIST_NEXT(lp, nfsl_list);
2121	lp->nfsl_seqid = 0;
2122	firstlock = 1;
2123	lop = LIST_FIRST(&lp->nfsl_lock);
2124	while (lop != NULL) {
2125	nlop = LIST_NEXT(lop, nfslo_list);
2126	if (lop->nfslo_end == NFS64BITSSET)
2127	len = NFS64BITSSET;
2128	else
2129	len = lop->nfslo_end - lop->nfslo_first;
2130	error = nfscl_trylock(nmp, NULL,
2131	op->nfso_fh, op->nfso_fhlen, lp,
2132	firstlock, 1, lop->nfslo_first, len,
2133	lop->nfslo_type, tcred, p);
2134	if (error != 0)
2135	nfscl_freelock(lop, 0);
2136	else
2137	firstlock = 0;
2138	lop = nlop;
2139	}
2140	/* If no locks, but a lockowner, just delete it. */
2141	if (LIST_EMPTY(&lp->nfsl_lock))
2142	nfscl_freelockowner(lp, 0);
2143	lp = nlp;
2144	}
2145	}
2146	}
2147	if (error != 0 && error != NFSERR_BADSESSION)
2148	nfscl_freeopen(op, 0);
2149	op = nop;
2150	}
2151	owp = nowp;
2152	}
2153
2154	/*
2155	* Now, try and get any delegations not yet reclaimed by cobbling
2156	* to-gether an appropriate open.
2157	*/
2158	nowp = NULL;
2159	dp = TAILQ_FIRST(&clp->nfsc_deleg);
2160	while (dp != NULL) {
2161	ndp = TAILQ_NEXT(dp, nfsdl_list);
2162	if ((dp->nfsdl_flags & NFSCLDL_NEEDRECLAIM)) {
2163	if (nowp == NULL) {
2164	nowp = malloc(
2165	sizeof (struct nfsclowner), M_NFSCLOWNER, M_WAITOK);
2166	/*
2167	* Name must be as long an largest possible
2168	* NFSV4CL_LOCKNAMELEN. 12 for now.
2169	*/
2170	NFSBCOPY("RECLAIMDELEG", nowp->nfsow_owner,
2171	NFSV4CL_LOCKNAMELEN);
2172	LIST_INIT(&nowp->nfsow_open);
2173	nowp->nfsow_clp = clp;
2174	nowp->nfsow_seqid = 0;
2175	nowp->nfsow_defunct = 0;
2176	nfscl_lockinit(&nowp->nfsow_rwlock);
2177	}
2178	nop = NULL;
2179	if (error != NFSERR_NOGRACE && error != NFSERR_BADSESSION) {
2180	nop = malloc(sizeof (struct nfsclopen) +
2181	dp->nfsdl_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
2182	nop->nfso_own = nowp;
2183	if ((dp->nfsdl_flags & NFSCLDL_WRITE)) {
2184	nop->nfso_mode = NFSV4OPEN_ACCESSWRITE;
2185	delegtype = NFSV4OPEN_DELEGATEWRITE;
2186	} else {
2187	nop->nfso_mode = NFSV4OPEN_ACCESSREAD;
2188	delegtype = NFSV4OPEN_DELEGATEREAD;
2189	}
2190	nop->nfso_opencnt = 0;
2191	nop->nfso_posixlock = 1;
2192	nop->nfso_fhlen = dp->nfsdl_fhlen;
2193	NFSBCOPY(dp->nfsdl_fh, nop->nfso_fh, dp->nfsdl_fhlen);
2194	LIST_INIT(&nop->nfso_lock);
2195	nop->nfso_stateid.seqid = 0;
2196	nop->nfso_stateid.other[0] = 0;
2197	nop->nfso_stateid.other[1] = 0;
2198	nop->nfso_stateid.other[2] = 0;
2199	newnfs_copycred(&dp->nfsdl_cred, tcred);
2200	newnfs_copyincred(tcred, &nop->nfso_cred);
2201	tdp = NULL;
2202	error = nfscl_tryopen(nmp, NULL, nop->nfso_fh,
2203	nop->nfso_fhlen, nop->nfso_fh, nop->nfso_fhlen,
2204	nop->nfso_mode, nop, NULL, 0, &tdp, 1,
2205	delegtype, tcred, p);
2206	if (tdp != NULL) {
2207	if ((tdp->nfsdl_flags & NFSCLDL_WRITE))
2208	mode = NFSV4OPEN_ACCESSWRITE;
2209	else
2210	mode = NFSV4OPEN_ACCESSREAD;
2211	if ((nop->nfso_mode & mode) == mode &&
2212	nop->nfso_fhlen == tdp->nfsdl_fhlen &&
2213	!NFSBCMP(nop->nfso_fh, tdp->nfsdl_fh,
2214	nop->nfso_fhlen)) {
2215	dp->nfsdl_stateid = tdp->nfsdl_stateid;
2216	dp->nfsdl_sizelimit = tdp->nfsdl_sizelimit;
2217	dp->nfsdl_ace = tdp->nfsdl_ace;
2218	dp->nfsdl_change = tdp->nfsdl_change;
2219	dp->nfsdl_flags &= ~NFSCLDL_NEEDRECLAIM;
2220	if ((tdp->nfsdl_flags & NFSCLDL_RECALL))
2221	dp->nfsdl_flags \|= NFSCLDL_RECALL;
2222	free(tdp, M_NFSCLDELEG);
2223	} else {
2224	TAILQ_INSERT_HEAD(&extra_deleg, tdp, nfsdl_list);
2225	}
2226	}
2227	}
2228	if (error) {
2229	if (nop != NULL)
2230	free(nop, M_NFSCLOPEN);
2231	/*
2232	* Couldn't reclaim it, so throw the state
2233	* away. Ouch!!
2234	*/
2235	nfscl_cleandeleg(dp);
2236	nfscl_freedeleg(&clp->nfsc_deleg, dp);
2237	} else {
2238	LIST_INSERT_HEAD(&extra_open, nop, nfso_list);
2239	}
2240	}
2241	dp = ndp;
2242	}
2243
2244	/*
2245	* Now, get rid of extra Opens and Delegations.
2246	*/
2247	LIST_FOREACH_SAFE(op, &extra_open, nfso_list, nop) {
2248	do {
2249	newnfs_copycred(&op->nfso_cred, tcred);
2250	error = nfscl_tryclose(op, tcred, nmp, p);
2251	if (error == NFSERR_GRACE)
2252	(void) nfs_catnap(PZERO, error, "nfsexcls");
2253	} while (error == NFSERR_GRACE);
2254	LIST_REMOVE(op, nfso_list);
2255	free(op, M_NFSCLOPEN);
2256	}
2257	if (nowp != NULL)
2258	free(nowp, M_NFSCLOWNER);
2259
2260	TAILQ_FOREACH_SAFE(dp, &extra_deleg, nfsdl_list, ndp) {
2261	do {
2262	newnfs_copycred(&dp->nfsdl_cred, tcred);
2263	error = nfscl_trydelegreturn(dp, tcred, nmp, p);
2264	if (error == NFSERR_GRACE)
2265	(void) nfs_catnap(PZERO, error, "nfsexdlg");
2266	} while (error == NFSERR_GRACE);
2267	TAILQ_REMOVE(&extra_deleg, dp, nfsdl_list);
2268	free(dp, M_NFSCLDELEG);
2269	}
2270
2271	/* For NFSv4.1 or later, do a RECLAIM_COMPLETE. */
2272	if (NFSHASNFSV4N(nmp))
2273	(void)nfsrpc_reclaimcomplete(nmp, cred, p);
2274
2275	NFSLOCKCLSTATE();
2276	clp->nfsc_flags &= ~NFSCLFLAGS_RECVRINPROG;
2277	wakeup(&clp->nfsc_flags);
2278	nfsv4_unlock(&clp->nfsc_lock, 0);
2279	NFSUNLOCKCLSTATE();
2280	NFSFREECRED(tcred);
2281	}
2282
2283	/*
2284	* This function is called when a server replies with NFSERR_EXPIRED.
2285	* It deletes all state for the client and does a fresh SetClientId/confirm.
2286	* XXX Someday it should post a signal to the process(es) that hold the
2287	* state, so they know that lock state has been lost.
2288	*/
2289	APPLESTATIC int
2290	nfscl_hasexpired(struct nfsclclient clp, u_int32_t clidrev, NFSPROC_T p)
2291	{
2292	struct nfsmount *nmp;
2293	struct ucred *cred;
2294	int igotlock = 0, error, trycnt;
2295
2296	/*
2297	* If the clientid has gone away or a new SetClientid has already
2298	* been done, just return ok.
2299	*/
2300	if (clp == NULL \|\| clidrev != clp->nfsc_clientidrev)
2301	return (0);
2302
2303	/*
2304	* First, lock the client structure, so everyone else will
2305	* block when trying to use state. Also, use NFSCLFLAGS_EXPIREIT so
2306	* that only one thread does the work.
2307	*/
2308	NFSLOCKCLSTATE();
2309	clp->nfsc_flags \|= NFSCLFLAGS_EXPIREIT;
2310	do {
2311	igotlock = nfsv4_lock(&clp->nfsc_lock, 1, NULL,
2312	NFSCLSTATEMUTEXPTR, NULL);
2313	} while (!igotlock && (clp->nfsc_flags & NFSCLFLAGS_EXPIREIT));
2314	if ((clp->nfsc_flags & NFSCLFLAGS_EXPIREIT) == 0) {
2315	if (igotlock)
2316	nfsv4_unlock(&clp->nfsc_lock, 0);
2317	NFSUNLOCKCLSTATE();
2318	return (0);
2319	}
2320	clp->nfsc_flags \|= NFSCLFLAGS_RECVRINPROG;
2321	NFSUNLOCKCLSTATE();
2322
2323	nmp = clp->nfsc_nmp;
2324	if (nmp == NULL)
2325	panic("nfscl expired");
2326	cred = newnfs_getcred();
2327	trycnt = 5;
2328	do {
2329	error = nfsrpc_setclient(nmp, clp, 0, cred, p);
2330	} while ((error == NFSERR_STALECLIENTID \|\|
2331	error == NFSERR_BADSESSION \|\|
2332	error == NFSERR_STALEDONTRECOVER) && --trycnt > 0);
2333	if (error) {
2334	NFSLOCKCLSTATE();
2335	clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2336	} else {
2337	/*
2338	* Expire the state for the client.
2339	*/
2340	nfscl_expireclient(clp, nmp, cred, p);
2341	NFSLOCKCLSTATE();
2342	clp->nfsc_flags \|= NFSCLFLAGS_HASCLIENTID;
2343	clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2344	}
2345	clp->nfsc_flags &= ~(NFSCLFLAGS_EXPIREIT \| NFSCLFLAGS_RECVRINPROG);
2346	wakeup(&clp->nfsc_flags);
2347	nfsv4_unlock(&clp->nfsc_lock, 0);
2348	NFSUNLOCKCLSTATE();
2349	NFSFREECRED(cred);
2350	return (error);
2351	}
2352
2353	/*
2354	* This function inserts a lock in the list after insert_lop.
2355	*/
2356	static void
2357	nfscl_insertlock(struct nfscllockowner lp, struct nfscllock new_lop,
2358	struct nfscllock *insert_lop, int local)
2359	{
2360
2361	if ((struct nfscllockowner *)insert_lop == lp)
2362	LIST_INSERT_HEAD(&lp->nfsl_lock, new_lop, nfslo_list);
2363	else
2364	LIST_INSERT_AFTER(insert_lop, new_lop, nfslo_list);
2365	if (local)
2366	nfsstatsv1.cllocallocks++;
2367	else
2368	nfsstatsv1.cllocks++;
2369	}
2370
2371	/*
2372	* This function updates the locking for a lock owner and given file. It
2373	* maintains a list of lock ranges ordered on increasing file offset that
2374	* are NFSCLLOCK_READ or NFSCLLOCK_WRITE and non-overlapping (aka POSIX style).
2375	* It always adds new_lop to the list and sometimes uses the one pointed
2376	* at by other_lopp.
2377	* Returns 1 if the locks were modified, 0 otherwise.
2378	*/
2379	static int
2380	nfscl_updatelock(struct nfscllockowner lp, struct nfscllock *new_lopp,
2381	struct nfscllock **other_lopp, int local)
2382	{
2383	struct nfscllock new_lop = new_lopp;
2384	struct nfscllock lop, tlop, *ilop;
2385	struct nfscllock *other_lop;
2386	int unlock = 0, modified = 0;
2387	u_int64_t tmp;
2388
2389	/*
2390	* Work down the list until the lock is merged.
2391	*/
2392	if (new_lop->nfslo_type == F_UNLCK)
2393	unlock = 1;
2394	ilop = (struct nfscllock *)lp;
2395	lop = LIST_FIRST(&lp->nfsl_lock);
2396	while (lop != NULL) {
2397	/*
2398	* Only check locks for this file that aren't before the start of
2399	* new lock's range.
2400	*/
2401	if (lop->nfslo_end >= new_lop->nfslo_first) {
2402	if (new_lop->nfslo_end < lop->nfslo_first) {
2403	/*
2404	* If the new lock ends before the start of the
2405	* current lock's range, no merge, just insert
2406	* the new lock.
2407	*/
2408	break;
2409	}
2410	if (new_lop->nfslo_type == lop->nfslo_type \|\|
2411	(new_lop->nfslo_first <= lop->nfslo_first &&
2412	new_lop->nfslo_end >= lop->nfslo_end)) {
2413	/*
2414	* This lock can be absorbed by the new lock/unlock.
2415	* This happens when it covers the entire range
2416	* of the old lock or is contiguous
2417	* with the old lock and is of the same type or an
2418	* unlock.
2419	*/
2420	if (new_lop->nfslo_type != lop->nfslo_type \|\|
2421	new_lop->nfslo_first != lop->nfslo_first \|\|
2422	new_lop->nfslo_end != lop->nfslo_end)
2423	modified = 1;
2424	if (lop->nfslo_first < new_lop->nfslo_first)
2425	new_lop->nfslo_first = lop->nfslo_first;
2426	if (lop->nfslo_end > new_lop->nfslo_end)
2427	new_lop->nfslo_end = lop->nfslo_end;
2428	tlop = lop;
2429	lop = LIST_NEXT(lop, nfslo_list);
2430	nfscl_freelock(tlop, local);
2431	continue;
2432	}
2433
2434	/*
2435	* All these cases are for contiguous locks that are not the
2436	* same type, so they can't be merged.
2437	*/
2438	if (new_lop->nfslo_first <= lop->nfslo_first) {
2439	/*
2440	* This case is where the new lock overlaps with the
2441	* first part of the old lock. Move the start of the
2442	* old lock to just past the end of the new lock. The
2443	* new lock will be inserted in front of the old, since
2444	* ilop hasn't been updated. (We are done now.)
2445	*/
2446	if (lop->nfslo_first != new_lop->nfslo_end) {
2447	lop->nfslo_first = new_lop->nfslo_end;
2448	modified = 1;
2449	}
2450	break;
2451	}
2452	if (new_lop->nfslo_end >= lop->nfslo_end) {
2453	/*
2454	* This case is where the new lock overlaps with the
2455	* end of the old lock's range. Move the old lock's
2456	* end to just before the new lock's first and insert
2457	* the new lock after the old lock.
2458	* Might not be done yet, since the new lock could
2459	* overlap further locks with higher ranges.
2460	*/
2461	if (lop->nfslo_end != new_lop->nfslo_first) {
2462	lop->nfslo_end = new_lop->nfslo_first;
2463	modified = 1;
2464	}
2465	ilop = lop;
2466	lop = LIST_NEXT(lop, nfslo_list);
2467	continue;
2468	}
2469	/*
2470	* The final case is where the new lock's range is in the
2471	* middle of the current lock's and splits the current lock
2472	* up. Use *other_lopp to handle the second part of the
2473	* split old lock range. (We are done now.)
2474	* For unlock, we use new_lop as other_lop and tmp, since
2475	* other_lop and new_lop are the same for this case.
2476	* We noted the unlock case above, so we don't need
2477	* new_lop->nfslo_type any longer.
2478	*/
2479	tmp = new_lop->nfslo_first;
2480	if (unlock) {
2481	other_lop = new_lop;
2482	*new_lopp = NULL;
2483	} else {
2484	other_lop = *other_lopp;
2485	*other_lopp = NULL;
2486	}
2487	other_lop->nfslo_first = new_lop->nfslo_end;
2488	other_lop->nfslo_end = lop->nfslo_end;
2489	other_lop->nfslo_type = lop->nfslo_type;
2490	lop->nfslo_end = tmp;
2491	nfscl_insertlock(lp, other_lop, lop, local);
2492	ilop = lop;
2493	modified = 1;
2494	break;
2495	}
2496	ilop = lop;
2497	lop = LIST_NEXT(lop, nfslo_list);
2498	if (lop == NULL)
2499	break;
2500	}
2501
2502	/*
2503	* Insert the new lock in the list at the appropriate place.
2504	*/
2505	if (!unlock) {
2506	nfscl_insertlock(lp, new_lop, ilop, local);
2507	*new_lopp = NULL;
2508	modified = 1;
2509	}
2510	return (modified);
2511	}
2512
2513	/*
2514	* This function must be run as a kernel thread.
2515	* It does Renew Ops and recovery, when required.
2516	*/
2517	APPLESTATIC void
2518	nfscl_renewthread(struct nfsclclient clp, NFSPROC_T p)
2519	{
2520	struct nfsclowner owp, nowp;
2521	struct nfsclopen *op;
2522	struct nfscllockowner lp, nlp;
2523	struct nfscldeleghead dh;
2524	struct nfscldeleg dp, ndp;
2525	struct ucred *cred;
2526	u_int32_t clidrev;
2527	int error, cbpathdown, islept, igotlock, ret, clearok;
2528	uint32_t recover_done_time = 0;
2529	time_t mytime;
2530	static time_t prevsec = 0;
2531	struct nfscllockownerfh lfhp, nlfhp;
2532	struct nfscllockownerfhhead lfh;
2533	struct nfscllayout lyp, nlyp;
2534	struct nfscldevinfo dip, ndip;
2535	struct nfscllayouthead rlh;
2536	struct nfsclrecalllayout *recallp;
2537	struct nfsclds *dsp;
2538
2539	cred = newnfs_getcred();
2540	NFSLOCKCLSTATE();
2541	clp->nfsc_flags \|= NFSCLFLAGS_HASTHREAD;
2542	NFSUNLOCKCLSTATE();
2543	for(;;) {
2544	newnfs_setroot(cred);
2545	cbpathdown = 0;
2546	if (clp->nfsc_flags & NFSCLFLAGS_RECOVER) {
2547	/*
2548	* Only allow one recover within 1/2 of the lease
2549	* duration (nfsc_renew).
2550	*/
2551	if (recover_done_time < NFSD_MONOSEC) {
2552	recover_done_time = NFSD_MONOSEC +
2553	clp->nfsc_renew;
2554	NFSCL_DEBUG(1, "Doing recovery..\n");
2555	nfscl_recover(clp, cred, p);
2556	} else {
2557	NFSCL_DEBUG(1, "Clear Recovery dt=%u ms=%jd\n",
2558	recover_done_time, (intmax_t)NFSD_MONOSEC);
2559	NFSLOCKCLSTATE();
2560	clp->nfsc_flags &= ~NFSCLFLAGS_RECOVER;
2561	NFSUNLOCKCLSTATE();
2562	}
2563	}
2564	if (clp->nfsc_expire <= NFSD_MONOSEC &&
2565	(clp->nfsc_flags & NFSCLFLAGS_HASCLIENTID)) {
2566	clp->nfsc_expire = NFSD_MONOSEC + clp->nfsc_renew;
2567	clidrev = clp->nfsc_clientidrev;
2568	error = nfsrpc_renew(clp, NULL, cred, p);
2569	if (error == NFSERR_CBPATHDOWN)
2570	cbpathdown = 1;
2571	else if (error == NFSERR_STALECLIENTID \|\|
2572	error == NFSERR_BADSESSION) {
2573	NFSLOCKCLSTATE();
2574	clp->nfsc_flags \|= NFSCLFLAGS_RECOVER;
2575	NFSUNLOCKCLSTATE();
2576	} else if (error == NFSERR_EXPIRED)
2577	(void) nfscl_hasexpired(clp, clidrev, p);
2578	}
2579
2580	checkdsrenew:
2581	if (NFSHASNFSV4N(clp->nfsc_nmp)) {
2582	/* Do renews for any DS sessions. */
2583	NFSLOCKMNT(clp->nfsc_nmp);
2584	/* Skip first entry, since the MDS is handled above. */
2585	dsp = TAILQ_FIRST(&clp->nfsc_nmp->nm_sess);
2586	if (dsp != NULL)
2587	dsp = TAILQ_NEXT(dsp, nfsclds_list);
2588	while (dsp != NULL) {
2589	if (dsp->nfsclds_expire <= NFSD_MONOSEC &&
2590	dsp->nfsclds_sess.nfsess_defunct == 0) {
2591	dsp->nfsclds_expire = NFSD_MONOSEC +
2592	clp->nfsc_renew;
2593	NFSUNLOCKMNT(clp->nfsc_nmp);
2594	(void)nfsrpc_renew(clp, dsp, cred, p);
2595	goto checkdsrenew;
2596	}
2597	dsp = TAILQ_NEXT(dsp, nfsclds_list);
2598	}
2599	NFSUNLOCKMNT(clp->nfsc_nmp);
2600	}
2601
2602	TAILQ_INIT(&dh);
2603	NFSLOCKCLSTATE();
2604	if (cbpathdown)
2605	/* It's a Total Recall! */
2606	nfscl_totalrecall(clp);
2607
2608	/*
2609	* Now, handle defunct owners.
2610	*/
2611	LIST_FOREACH_SAFE(owp, &clp->nfsc_owner, nfsow_list, nowp) {
2612	if (LIST_EMPTY(&owp->nfsow_open)) {
2613	if (owp->nfsow_defunct != 0)
2614	nfscl_freeopenowner(owp, 0);
2615	}
2616	}
2617
2618	/*
2619	* Do the recall on any delegations. To avoid trouble, always
2620	* come back up here after having slept.
2621	*/
2622	igotlock = 0;
2623	tryagain:
2624	dp = TAILQ_FIRST(&clp->nfsc_deleg);
2625	while (dp != NULL) {
2626	ndp = TAILQ_NEXT(dp, nfsdl_list);
2627	if ((dp->nfsdl_flags & NFSCLDL_RECALL)) {
2628	/*
2629	* Wait for outstanding I/O ops to be done.
2630	*/
2631	if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
2632	if (igotlock) {
2633	nfsv4_unlock(&clp->nfsc_lock, 0);
2634	igotlock = 0;
2635	}
2636	dp->nfsdl_rwlock.nfslock_lock \|=
2637	NFSV4LOCK_WANTED;
2638	(void) nfsmsleep(&dp->nfsdl_rwlock,
2639	NFSCLSTATEMUTEXPTR, PZERO, "nfscld",
2640	NULL);
2641	goto tryagain;
2642	}
2643	while (!igotlock) {
2644	igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
2645	&islept, NFSCLSTATEMUTEXPTR, NULL);
2646	if (islept)
2647	goto tryagain;
2648	}
2649	NFSUNLOCKCLSTATE();
2650	newnfs_copycred(&dp->nfsdl_cred, cred);
2651	ret = nfscl_recalldeleg(clp, clp->nfsc_nmp, dp,
2652	NULL, cred, p, 1);
2653	if (!ret) {
2654	nfscl_cleandeleg(dp);
2655	TAILQ_REMOVE(&clp->nfsc_deleg, dp,
2656	nfsdl_list);
2657	LIST_REMOVE(dp, nfsdl_hash);
2658	TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2659	nfscl_delegcnt--;
2660	nfsstatsv1.cldelegates--;
2661	}
2662	NFSLOCKCLSTATE();
2663	}
2664	dp = ndp;
2665	}
2666
2667	/*
2668	* Clear out old delegations, if we are above the high water
2669	* mark. Only clear out ones with no state related to them.
2670	* The tailq list is in LRU order.
2671	*/
2672	dp = TAILQ_LAST(&clp->nfsc_deleg, nfscldeleghead);
2673	while (nfscl_delegcnt > nfscl_deleghighwater && dp != NULL) {
2674	ndp = TAILQ_PREV(dp, nfscldeleghead, nfsdl_list);
2675	if (dp->nfsdl_rwlock.nfslock_usecnt == 0 &&
2676	dp->nfsdl_rwlock.nfslock_lock == 0 &&
2677	dp->nfsdl_timestamp < NFSD_MONOSEC &&
2678	(dp->nfsdl_flags & (NFSCLDL_RECALL \| NFSCLDL_ZAPPED \|
2679	NFSCLDL_NEEDRECLAIM \| NFSCLDL_DELEGRET)) == 0) {
2680	clearok = 1;
2681	LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2682	op = LIST_FIRST(&owp->nfsow_open);
2683	if (op != NULL) {
2684	clearok = 0;
2685	break;
2686	}
2687	}
2688	if (clearok) {
2689	LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
2690	if (!LIST_EMPTY(&lp->nfsl_lock)) {
2691	clearok = 0;
2692	break;
2693	}
2694	}
2695	}
2696	if (clearok) {
2697	TAILQ_REMOVE(&clp->nfsc_deleg, dp, nfsdl_list);
2698	LIST_REMOVE(dp, nfsdl_hash);
2699	TAILQ_INSERT_HEAD(&dh, dp, nfsdl_list);
2700	nfscl_delegcnt--;
2701	nfsstatsv1.cldelegates--;
2702	}
2703	}
2704	dp = ndp;
2705	}
2706	if (igotlock)
2707	nfsv4_unlock(&clp->nfsc_lock, 0);
2708
2709	/*
2710	* Do the recall on any layouts. To avoid trouble, always
2711	* come back up here after having slept.
2712	*/
2713	TAILQ_INIT(&rlh);
2714	tryagain2:
2715	TAILQ_FOREACH_SAFE(lyp, &clp->nfsc_layout, nfsly_list, nlyp) {
2716	if ((lyp->nfsly_flags & NFSLY_RECALL) != 0) {
2717	/*
2718	* Wait for outstanding I/O ops to be done.
2719	*/
2720	if (lyp->nfsly_lock.nfslock_usecnt > 0 \|\|
2721	(lyp->nfsly_lock.nfslock_lock &
2722	NFSV4LOCK_LOCK) != 0) {
2723	lyp->nfsly_lock.nfslock_lock \|=
2724	NFSV4LOCK_WANTED;
2725	nfsmsleep(&lyp->nfsly_lock.nfslock_lock,
2726	NFSCLSTATEMUTEXPTR, PZERO, "nfslyp",
2727	NULL);
2728	goto tryagain2;
2729	}
2730	/* Move the layout to the recall list. */
2731	TAILQ_REMOVE(&clp->nfsc_layout, lyp,
2732	nfsly_list);
2733	LIST_REMOVE(lyp, nfsly_hash);
2734	TAILQ_INSERT_HEAD(&rlh, lyp, nfsly_list);
2735
2736	/* Handle any layout commits. */
2737	if (!NFSHASNOLAYOUTCOMMIT(clp->nfsc_nmp) &&
2738	(lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
2739	lyp->nfsly_flags &= ~NFSLY_WRITTEN;
2740	NFSUNLOCKCLSTATE();
2741	NFSCL_DEBUG(3, "do layoutcommit\n");
2742	nfscl_dolayoutcommit(clp->nfsc_nmp, lyp,
2743	cred, p);
2744	NFSLOCKCLSTATE();
2745	goto tryagain2;
2746	}
2747	}
2748	}
2749
2750	/* Now, look for stale layouts. */
2751	lyp = TAILQ_LAST(&clp->nfsc_layout, nfscllayouthead);
2752	while (lyp != NULL) {
2753	nlyp = TAILQ_PREV(lyp, nfscllayouthead, nfsly_list);
2754	if (lyp->nfsly_timestamp < NFSD_MONOSEC &&
2755	(lyp->nfsly_flags & NFSLY_RECALL) == 0 &&
2756	lyp->nfsly_lock.nfslock_usecnt == 0 &&
2757	lyp->nfsly_lock.nfslock_lock == 0) {
2758	NFSCL_DEBUG(4, "ret stale lay=%d\n",
2759	nfscl_layoutcnt);
2760	recallp = malloc(sizeof(*recallp),
2761	M_NFSLAYRECALL, M_NOWAIT);
2762	if (recallp == NULL)
2763	break;
2764	(void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE,
2765	lyp, NFSLAYOUTIOMODE_ANY, 0, UINT64_MAX,
2766	lyp->nfsly_stateid.seqid, 0, 0, NULL,
2767	recallp);
2768	}
2769	lyp = nlyp;
2770	}
2771
2772	/*
2773	* Free up any unreferenced device info structures.
2774	*/
2775	LIST_FOREACH_SAFE(dip, &clp->nfsc_devinfo, nfsdi_list, ndip) {
2776	if (dip->nfsdi_layoutrefs == 0 &&
2777	dip->nfsdi_refcnt == 0) {
2778	NFSCL_DEBUG(4, "freeing devinfo\n");
2779	LIST_REMOVE(dip, nfsdi_list);
2780	nfscl_freedevinfo(dip);
2781	}
2782	}
2783	NFSUNLOCKCLSTATE();
2784
2785	/* Do layout return(s), as required. */
2786	TAILQ_FOREACH_SAFE(lyp, &rlh, nfsly_list, nlyp) {
2787	TAILQ_REMOVE(&rlh, lyp, nfsly_list);
2788	NFSCL_DEBUG(4, "ret layout\n");
2789	nfscl_layoutreturn(clp->nfsc_nmp, lyp, cred, p);
2790	nfscl_freelayout(lyp);
2791	}
2792
2793	/*
2794	* Delegreturn any delegations cleaned out or recalled.
2795	*/
2796	TAILQ_FOREACH_SAFE(dp, &dh, nfsdl_list, ndp) {
2797	newnfs_copycred(&dp->nfsdl_cred, cred);
2798	(void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
2799	TAILQ_REMOVE(&dh, dp, nfsdl_list);
2800	free(dp, M_NFSCLDELEG);
2801	}
2802
2803	SLIST_INIT(&lfh);
2804	/*
2805	* Call nfscl_cleanupkext() once per second to check for
2806	* open/lock owners where the process has exited.
2807	*/
2808	mytime = NFSD_MONOSEC;
2809	if (prevsec != mytime) {
2810	prevsec = mytime;
2811	nfscl_cleanupkext(clp, &lfh);
2812	}
2813
2814	/*
2815	* Do a ReleaseLockOwner for all lock owners where the
2816	* associated process no longer exists, as found by
2817	* nfscl_cleanupkext().
2818	*/
2819	newnfs_setroot(cred);
2820	SLIST_FOREACH_SAFE(lfhp, &lfh, nfslfh_list, nlfhp) {
2821	LIST_FOREACH_SAFE(lp, &lfhp->nfslfh_lock, nfsl_list,
2822	nlp) {
2823	(void)nfsrpc_rellockown(clp->nfsc_nmp, lp,
2824	lfhp->nfslfh_fh, lfhp->nfslfh_len, cred,
2825	p);
2826	nfscl_freelockowner(lp, 0);
2827	}
2828	free(lfhp, M_TEMP);
2829	}
2830	SLIST_INIT(&lfh);
2831
2832	NFSLOCKCLSTATE();
2833	if ((clp->nfsc_flags & NFSCLFLAGS_RECOVER) == 0)
2834	(void)mtx_sleep(clp, NFSCLSTATEMUTEXPTR, PWAIT, "nfscl",
2835	hz);
2836	if (clp->nfsc_flags & NFSCLFLAGS_UMOUNT) {
2837	clp->nfsc_flags &= ~NFSCLFLAGS_HASTHREAD;
2838	NFSUNLOCKCLSTATE();
2839	NFSFREECRED(cred);
2840	wakeup((caddr_t)clp);
2841	return;
2842	}
2843	NFSUNLOCKCLSTATE();
2844	}
2845	}
2846
2847	/*
2848	* Initiate state recovery. Called when NFSERR_STALECLIENTID,
2849	* NFSERR_STALESTATEID or NFSERR_BADSESSION is received.
2850	*/
2851	APPLESTATIC void
2852	nfscl_initiate_recovery(struct nfsclclient *clp)
2853	{
2854
2855	if (clp == NULL)
2856	return;
2857	NFSLOCKCLSTATE();
2858	clp->nfsc_flags \|= NFSCLFLAGS_RECOVER;
2859	NFSUNLOCKCLSTATE();
2860	wakeup((caddr_t)clp);
2861	}
2862
2863	/*
2864	* Dump out the state stuff for debugging.
2865	*/
2866	APPLESTATIC void
2867	nfscl_dumpstate(struct nfsmount *nmp, int openowner, int opens,
2868	int lockowner, int locks)
2869	{
2870	struct nfsclclient *clp;
2871	struct nfsclowner *owp;
2872	struct nfsclopen *op;
2873	struct nfscllockowner *lp;
2874	struct nfscllock *lop;
2875	struct nfscldeleg *dp;
2876
2877	clp = nmp->nm_clp;
2878	if (clp == NULL) {
2879	printf("nfscl dumpstate NULL clp\n");
2880	return;
2881	}
2882	NFSLOCKCLSTATE();
2883	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
2884	LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
2885	if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2886	printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2887	owp->nfsow_owner[0], owp->nfsow_owner[1],
2888	owp->nfsow_owner[2], owp->nfsow_owner[3],
2889	owp->nfsow_seqid);
2890	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2891	if (opens)
2892	printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2893	op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2894	op->nfso_stateid.other[2], op->nfso_opencnt,
2895	op->nfso_fh[12]);
2896	LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2897	if (lockowner)
2898	printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2899	lp->nfsl_owner[0], lp->nfsl_owner[1],
2900	lp->nfsl_owner[2], lp->nfsl_owner[3],
2901	lp->nfsl_seqid,
2902	lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2903	lp->nfsl_stateid.other[2]);
2904	LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2905	if (locks)
2906	#ifdef __FreeBSD__
2907	printf("lck typ=%d fst=%ju end=%ju\n",
2908	lop->nfslo_type, (intmax_t)lop->nfslo_first,
2909	(intmax_t)lop->nfslo_end);
2910	#else
2911	printf("lck typ=%d fst=%qd end=%qd\n",
2912	lop->nfslo_type, lop->nfslo_first,
2913	lop->nfslo_end);
2914	#endif
2915	}
2916	}
2917	}
2918	}
2919	}
2920	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2921	if (openowner && !LIST_EMPTY(&owp->nfsow_open))
2922	printf("owner=0x%x 0x%x 0x%x 0x%x seqid=%d\n",
2923	owp->nfsow_owner[0], owp->nfsow_owner[1],
2924	owp->nfsow_owner[2], owp->nfsow_owner[3],
2925	owp->nfsow_seqid);
2926	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
2927	if (opens)
2928	printf("open st=0x%x 0x%x 0x%x cnt=%d fh12=0x%x\n",
2929	op->nfso_stateid.other[0], op->nfso_stateid.other[1],
2930	op->nfso_stateid.other[2], op->nfso_opencnt,
2931	op->nfso_fh[12]);
2932	LIST_FOREACH(lp, &op->nfso_lock, nfsl_list) {
2933	if (lockowner)
2934	printf("lckown=0x%x 0x%x 0x%x 0x%x seqid=%d st=0x%x 0x%x 0x%x\n",
2935	lp->nfsl_owner[0], lp->nfsl_owner[1],
2936	lp->nfsl_owner[2], lp->nfsl_owner[3],
2937	lp->nfsl_seqid,
2938	lp->nfsl_stateid.other[0], lp->nfsl_stateid.other[1],
2939	lp->nfsl_stateid.other[2]);
2940	LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
2941	if (locks)
2942	#ifdef __FreeBSD__
2943	printf("lck typ=%d fst=%ju end=%ju\n",
2944	lop->nfslo_type, (intmax_t)lop->nfslo_first,
2945	(intmax_t)lop->nfslo_end);
2946	#else
2947	printf("lck typ=%d fst=%qd end=%qd\n",
2948	lop->nfslo_type, lop->nfslo_first,
2949	lop->nfslo_end);
2950	#endif
2951	}
2952	}
2953	}
2954	}
2955	NFSUNLOCKCLSTATE();
2956	}
2957
2958	/*
2959	* Check for duplicate open owners and opens.
2960	* (Only used as a diagnostic aid.)
2961	*/
2962	APPLESTATIC void
2963	nfscl_dupopen(vnode_t vp, int dupopens)
2964	{
2965	struct nfsclclient *clp;
2966	struct nfsclowner owp, owp2;
2967	struct nfsclopen op, op2;
2968	struct nfsfh *nfhp;
2969
2970	clp = VFSTONFS(vnode_mount(vp))->nm_clp;
2971	if (clp == NULL) {
2972	printf("nfscl dupopen NULL clp\n");
2973	return;
2974	}
2975	nfhp = VTONFS(vp)->n_fhp;
2976	NFSLOCKCLSTATE();
2977
2978	/*
2979	* First, search for duplicate owners.
2980	* These should never happen!
2981	*/
2982	LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
2983	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
2984	if (owp != owp2 &&
2985	!NFSBCMP(owp->nfsow_owner, owp2->nfsow_owner,
2986	NFSV4CL_LOCKNAMELEN)) {
2987	NFSUNLOCKCLSTATE();
2988	printf("DUP OWNER\n");
2989	nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0, 0);
2990	return;
2991	}
2992	}
2993	}
2994
2995	/*
2996	* Now, search for duplicate stateids.
2997	* These shouldn't happen, either.
2998	*/
2999	LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3000	LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3001	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3002	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3003	if (op != op2 &&
3004	(op->nfso_stateid.other[0] != 0 \|\|
3005	op->nfso_stateid.other[1] != 0 \|\|
3006	op->nfso_stateid.other[2] != 0) &&
3007	op->nfso_stateid.other[0] == op2->nfso_stateid.other[0] &&
3008	op->nfso_stateid.other[1] == op2->nfso_stateid.other[1] &&
3009	op->nfso_stateid.other[2] == op2->nfso_stateid.other[2]) {
3010	NFSUNLOCKCLSTATE();
3011	printf("DUP STATEID\n");
3012	nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1, 0,
3013	0);
3014	return;
3015	}
3016	}
3017	}
3018	}
3019	}
3020
3021	/*
3022	* Now search for duplicate opens.
3023	* Duplicate opens for the same owner
3024	* should never occur. Other duplicates are
3025	* possible and are checked for if "dupopens"
3026	* is true.
3027	*/
3028	LIST_FOREACH(owp2, &clp->nfsc_owner, nfsow_list) {
3029	LIST_FOREACH(op2, &owp2->nfsow_open, nfso_list) {
3030	if (nfhp->nfh_len == op2->nfso_fhlen &&
3031	!NFSBCMP(nfhp->nfh_fh, op2->nfso_fh, nfhp->nfh_len)) {
3032	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3033	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3034	if (op != op2 && nfhp->nfh_len == op->nfso_fhlen &&
3035	!NFSBCMP(nfhp->nfh_fh, op->nfso_fh, nfhp->nfh_len) &&
3036	(!NFSBCMP(op->nfso_own->nfsow_owner,
3037	op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN) \|\|
3038	dupopens)) {
3039	if (!NFSBCMP(op->nfso_own->nfsow_owner,
3040	op2->nfso_own->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3041	NFSUNLOCKCLSTATE();
3042	printf("BADDUP OPEN\n");
3043	} else {
3044	NFSUNLOCKCLSTATE();
3045	printf("DUP OPEN\n");
3046	}
3047	nfscl_dumpstate(VFSTONFS(vnode_mount(vp)), 1, 1,
3048	0, 0);
3049	return;
3050	}
3051	}
3052	}
3053	}
3054	}
3055	}
3056	NFSUNLOCKCLSTATE();
3057	}
3058
3059	/*
3060	* During close, find an open that needs to be dereferenced and
3061	* dereference it. If there are no more opens for this file,
3062	* log a message to that effect.
3063	* Opens aren't actually Close'd until VOP_INACTIVE() is performed
3064	* on the file's vnode.
3065	* This is the safe way, since it is difficult to identify
3066	* which open the close is for and I/O can be performed after the
3067	* close(2) system call when a file is mmap'd.
3068	* If it returns 0 for success, there will be a referenced
3069	* clp returned via clpp.
3070	*/
3071	APPLESTATIC int
3072	nfscl_getclose(vnode_t vp, struct nfsclclient **clpp)
3073	{
3074	struct nfsclclient *clp;
3075	struct nfsclowner *owp;
3076	struct nfsclopen *op;
3077	struct nfscldeleg *dp;
3078	struct nfsfh *nfhp;
3079	int error, notdecr;
3080
3081	error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp);
3082	if (error)
3083	return (error);
3084	*clpp = clp;
3085
3086	nfhp = VTONFS(vp)->n_fhp;
3087	notdecr = 1;
3088	NFSLOCKCLSTATE();
3089	/*
3090	* First, look for one under a delegation that was locally issued
3091	* and just decrement the opencnt for it. Since all my Opens against
3092	* the server are DENY_NONE, I don't see a problem with hanging
3093	* onto them. (It is much easier to use one of the extant Opens
3094	* that I already have on the server when a Delegation is recalled
3095	* than to do fresh Opens.) Someday, I might need to rethink this, but.
3096	*/
3097	dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3098	if (dp != NULL) {
3099	LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
3100	op = LIST_FIRST(&owp->nfsow_open);
3101	if (op != NULL) {
3102	/*
3103	* Since a delegation is for a file, there
3104	* should never be more than one open for
3105	* each openowner.
3106	*/
3107	if (LIST_NEXT(op, nfso_list) != NULL)
3108	panic("nfscdeleg opens");
3109	if (notdecr && op->nfso_opencnt > 0) {
3110	notdecr = 0;
3111	op->nfso_opencnt--;
3112	break;
3113	}
3114	}
3115	}
3116	}
3117
3118	/* Now process the opens against the server. */
3119	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3120	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3121	if (op->nfso_fhlen == nfhp->nfh_len &&
3122	!NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3123	nfhp->nfh_len)) {
3124	/* Found an open, decrement cnt if possible */
3125	if (notdecr && op->nfso_opencnt > 0) {
3126	notdecr = 0;
3127	op->nfso_opencnt--;
3128	}
3129	/*
3130	* There are more opens, so just return.
3131	*/
3132	if (op->nfso_opencnt > 0) {
3133	NFSUNLOCKCLSTATE();
3134	return (0);
3135	}
3136	}
3137	}
3138	}
3139	NFSUNLOCKCLSTATE();
3140	if (notdecr)
3141	printf("nfscl: never fnd open\n");
3142	return (0);
3143	}
3144
3145	APPLESTATIC int
3146	nfscl_doclose(vnode_t vp, struct nfsclclient *clpp, NFSPROC_T p)
3147	{
3148	struct nfsclclient *clp;
3149	struct nfsclowner owp, nowp;
3150	struct nfsclopen *op;
3151	struct nfscldeleg *dp;
3152	struct nfsfh *nfhp;
3153	struct nfsclrecalllayout *recallp;
3154	int error;
3155
3156	error = nfscl_getcl(vnode_mount(vp), NULL, NULL, 1, &clp);
3157	if (error)
3158	return (error);
3159	*clpp = clp;
3160
3161	nfhp = VTONFS(vp)->n_fhp;
3162	recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
3163	NFSLOCKCLSTATE();
3164	/*
3165	* First get rid of the local Open structures, which should be no
3166	* longer in use.
3167	*/
3168	dp = nfscl_finddeleg(clp, nfhp->nfh_fh, nfhp->nfh_len);
3169	if (dp != NULL) {
3170	LIST_FOREACH_SAFE(owp, &dp->nfsdl_owner, nfsow_list, nowp) {
3171	op = LIST_FIRST(&owp->nfsow_open);
3172	if (op != NULL) {
3173	KASSERT((op->nfso_opencnt == 0),
3174	("nfscl: bad open cnt on deleg"));
3175	nfscl_freeopen(op, 1);
3176	}
3177	nfscl_freeopenowner(owp, 1);
3178	}
3179	}
3180
3181	/* Return any layouts marked return on close. */
3182	nfscl_retoncloselayout(vp, clp, nfhp->nfh_fh, nfhp->nfh_len, &recallp);
3183
3184	/* Now process the opens against the server. */
3185	lookformore:
3186	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3187	op = LIST_FIRST(&owp->nfsow_open);
3188	while (op != NULL) {
3189	if (op->nfso_fhlen == nfhp->nfh_len &&
3190	!NFSBCMP(op->nfso_fh, nfhp->nfh_fh,
3191	nfhp->nfh_len)) {
3192	/* Found an open, close it. */
3193	KASSERT((op->nfso_opencnt == 0),
3194	("nfscl: bad open cnt on server"));
3195	NFSUNLOCKCLSTATE();
3196	nfsrpc_doclose(VFSTONFS(vnode_mount(vp)), op,
3197	p);
3198	NFSLOCKCLSTATE();
3199	goto lookformore;
3200	}
3201	op = LIST_NEXT(op, nfso_list);
3202	}
3203	}
3204	NFSUNLOCKCLSTATE();
3205	/*
3206	* recallp has been set NULL by nfscl_retoncloselayout() if it was
3207	* used by the function, but calling free() with a NULL pointer is ok.
3208	*/
3209	free(recallp, M_NFSLAYRECALL);
3210	return (0);
3211	}
3212
3213	/*
3214	* Return all delegations on this client.
3215	* (Must be called with client sleep lock.)
3216	*/
3217	static void
3218	nfscl_delegreturnall(struct nfsclclient clp, NFSPROC_T p)
3219	{
3220	struct nfscldeleg dp, ndp;
3221	struct ucred *cred;
3222
3223	cred = newnfs_getcred();
3224	TAILQ_FOREACH_SAFE(dp, &clp->nfsc_deleg, nfsdl_list, ndp) {
3225	nfscl_cleandeleg(dp);
3226	(void) nfscl_trydelegreturn(dp, cred, clp->nfsc_nmp, p);
3227	nfscl_freedeleg(&clp->nfsc_deleg, dp);
3228	}
3229	NFSFREECRED(cred);
3230	}
3231
3232	/*
3233	* Do a callback RPC.
3234	*/
3235	APPLESTATIC void
3236	nfscl_docb(struct nfsrv_descript nd, NFSPROC_T p)
3237	{
3238	int clist, gotseq_ok, i, j, k, op, rcalls;
3239	u_int32_t *tl;
3240	struct nfsclclient *clp;
3241	struct nfscldeleg *dp = NULL;
3242	int numops, taglen = -1, error = 0, trunc __unused;
3243	u_int32_t minorvers = 0, retops = 0, retopsp = NULL, repp, cbident;
3244	u_char tag[NFSV4_SMALLSTR + 1], *tagstr;
3245	vnode_t vp = NULL;
3246	struct nfsnode *np;
3247	struct vattr va;
3248	struct nfsfh *nfhp;
3249	mount_t mp;
3250	nfsattrbit_t attrbits, rattrbits;
3251	nfsv4stateid_t stateid;
3252	uint32_t seqid, slotid = 0, highslot, cachethis __unused;
3253	uint8_t sessionid[NFSX_V4SESSIONID];
3254	struct mbuf *rep;
3255	struct nfscllayout *lyp;
3256	uint64_t filesid[2], len, off;
3257	int changed, gotone, laytype, recalltype;
3258	uint32_t iomode;
3259	struct nfsclrecalllayout *recallp = NULL;
3260	struct nfsclsession *tsep;
3261
3262	gotseq_ok = 0;
3263	nfsrvd_rephead(nd);
3264	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3265	taglen = fxdr_unsigned(int, *tl);
3266	if (taglen < 0) {
3267	error = EBADRPC;
3268	goto nfsmout;
3269	}
3270	if (taglen <= NFSV4_SMALLSTR)
3271	tagstr = tag;
3272	else
3273	tagstr = malloc(taglen + 1, M_TEMP, M_WAITOK);
3274	error = nfsrv_mtostr(nd, tagstr, taglen);
3275	if (error) {
3276	if (taglen > NFSV4_SMALLSTR)
3277	free(tagstr, M_TEMP);
3278	taglen = -1;
3279	goto nfsmout;
3280	}
3281	(void) nfsm_strtom(nd, tag, taglen);
3282	if (taglen > NFSV4_SMALLSTR) {
3283	free(tagstr, M_TEMP);
3284	}
3285	NFSM_BUILD(retopsp, u_int32_t *, NFSX_UNSIGNED);
3286	NFSM_DISSECT(tl, u_int32_t , 3 NFSX_UNSIGNED);
3287	minorvers = fxdr_unsigned(u_int32_t, *tl++);
3288	if (minorvers != NFSV4_MINORVERSION && minorvers != NFSV41_MINORVERSION)
3289	nd->nd_repstat = NFSERR_MINORVERMISMATCH;
3290	cbident = fxdr_unsigned(u_int32_t, *tl++);
3291	if (nd->nd_repstat)
3292	numops = 0;
3293	else
3294	numops = fxdr_unsigned(int, *tl);
3295	/*
3296	* Loop around doing the sub ops.
3297	*/
3298	for (i = 0; i < numops; i++) {
3299	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3300	NFSM_BUILD(repp, u_int32_t , 2 NFSX_UNSIGNED);
3301	repp++ = tl;
3302	op = fxdr_unsigned(int, *tl);
3303	if (op < NFSV4OP_CBGETATTR \|\|
3304	(op > NFSV4OP_CBRECALL && minorvers == NFSV4_MINORVERSION) \|\|
3305	(op > NFSV4OP_CBNOTIFYDEVID &&
3306	minorvers == NFSV41_MINORVERSION)) {
3307	nd->nd_repstat = NFSERR_OPILLEGAL;
3308	*repp = nfscl_errmap(nd, minorvers);
3309	retops++;
3310	break;
3311	}
3312	nd->nd_procnum = op;
3313	if (op < NFSV41_CBNOPS)
3314	nfsstatsv1.cbrpccnt[nd->nd_procnum]++;
3315	switch (op) {
3316	case NFSV4OP_CBGETATTR:
3317	NFSCL_DEBUG(4, "cbgetattr\n");
3318	mp = NULL;
3319	vp = NULL;
3320	error = nfsm_getfh(nd, &nfhp);
3321	if (!error)
3322	error = nfsrv_getattrbits(nd, &attrbits,
3323	NULL, NULL);
3324	if (error == 0 && i == 0 &&
3325	minorvers != NFSV4_MINORVERSION)
3326	error = NFSERR_OPNOTINSESS;
3327	if (!error) {
3328	mp = nfscl_getmnt(minorvers, sessionid, cbident,
3329	&clp);
3330	if (mp == NULL)
3331	error = NFSERR_SERVERFAULT;
3332	}
3333	if (!error) {
3334	error = nfscl_ngetreopen(mp, nfhp->nfh_fh,
3335	nfhp->nfh_len, p, &np);
3336	if (!error)
3337	vp = NFSTOV(np);
3338	}
3339	if (!error) {
3340	NFSZERO_ATTRBIT(&rattrbits);
3341	NFSLOCKCLSTATE();
3342	dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3343	nfhp->nfh_len);
3344	if (dp != NULL) {
3345	if (NFSISSET_ATTRBIT(&attrbits,
3346	NFSATTRBIT_SIZE)) {
3347	if (vp != NULL)
3348	va.va_size = np->n_size;
3349	else
3350	va.va_size =
3351	dp->nfsdl_size;
3352	NFSSETBIT_ATTRBIT(&rattrbits,
3353	NFSATTRBIT_SIZE);
3354	}
3355	if (NFSISSET_ATTRBIT(&attrbits,
3356	NFSATTRBIT_CHANGE)) {
3357	va.va_filerev =
3358	dp->nfsdl_change;
3359	if (vp == NULL \|\|
3360	(np->n_flag & NDELEGMOD))
3361	va.va_filerev++;
3362	NFSSETBIT_ATTRBIT(&rattrbits,
3363	NFSATTRBIT_CHANGE);
3364	}
3365	} else
3366	error = NFSERR_SERVERFAULT;
3367	NFSUNLOCKCLSTATE();
3368	}
3369	if (vp != NULL)
3370	vrele(vp);
3371	if (mp != NULL)
3372	vfs_unbusy(mp);
3373	if (nfhp != NULL)
3374	free(nfhp, M_NFSFH);
3375	if (!error)
3376	(void) nfsv4_fillattr(nd, NULL, NULL, NULL, &va,
3377	NULL, 0, &rattrbits, NULL, p, 0, 0, 0, 0,
3378	(uint64_t)0, NULL);
3379	break;
3380	case NFSV4OP_CBRECALL:
3381	NFSCL_DEBUG(4, "cbrecall\n");
3382	NFSM_DISSECT(tl, u_int32_t *, NFSX_STATEID +
3383	NFSX_UNSIGNED);
3384	stateid.seqid = *tl++;
3385	NFSBCOPY((caddr_t)tl, (caddr_t)stateid.other,
3386	NFSX_STATEIDOTHER);
3387	tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3388	trunc = fxdr_unsigned(int, *tl);
3389	error = nfsm_getfh(nd, &nfhp);
3390	if (error == 0 && i == 0 &&
3391	minorvers != NFSV4_MINORVERSION)
3392	error = NFSERR_OPNOTINSESS;
3393	if (!error) {
3394	NFSLOCKCLSTATE();
3395	if (minorvers == NFSV4_MINORVERSION)
3396	clp = nfscl_getclnt(cbident);
3397	else
3398	clp = nfscl_getclntsess(sessionid);
3399	if (clp != NULL) {
3400	dp = nfscl_finddeleg(clp, nfhp->nfh_fh,
3401	nfhp->nfh_len);
3402	if (dp != NULL && (dp->nfsdl_flags &
3403	NFSCLDL_DELEGRET) == 0) {
3404	dp->nfsdl_flags \|=
3405	NFSCLDL_RECALL;
3406	wakeup((caddr_t)clp);
3407	}
3408	} else {
3409	error = NFSERR_SERVERFAULT;
3410	}
3411	NFSUNLOCKCLSTATE();
3412	}
3413	if (nfhp != NULL)
3414	free(nfhp, M_NFSFH);
3415	break;
3416	case NFSV4OP_CBLAYOUTRECALL:
3417	NFSCL_DEBUG(4, "cblayrec\n");
3418	nfhp = NULL;
3419	NFSM_DISSECT(tl, uint32_t , 4 NFSX_UNSIGNED);
3420	laytype = fxdr_unsigned(int, *tl++);
3421	iomode = fxdr_unsigned(uint32_t, *tl++);
3422	if (newnfs_true == *tl++)
3423	changed = 1;
3424	else
3425	changed = 0;
3426	recalltype = fxdr_unsigned(int, *tl);
3427	NFSCL_DEBUG(4, "layt=%d iom=%d ch=%d rectyp=%d\n",
3428	laytype, iomode, changed, recalltype);
3429	recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL,
3430	M_WAITOK);
3431	if (laytype != NFSLAYOUT_NFSV4_1_FILES &&
3432	laytype != NFSLAYOUT_FLEXFILE)
3433	error = NFSERR_NOMATCHLAYOUT;
3434	else if (recalltype == NFSLAYOUTRETURN_FILE) {
3435	error = nfsm_getfh(nd, &nfhp);
3436	NFSCL_DEBUG(4, "retfile getfh=%d\n", error);
3437	if (error != 0)
3438	goto nfsmout;
3439	NFSM_DISSECT(tl, u_int32_t , 2 NFSX_HYPER +
3440	NFSX_STATEID);
3441	off = fxdr_hyper(tl); tl += 2;
3442	len = fxdr_hyper(tl); tl += 2;
3443	stateid.seqid = fxdr_unsigned(uint32_t, *tl++);
3444	NFSBCOPY(tl, stateid.other, NFSX_STATEIDOTHER);
3445	if (minorvers == NFSV4_MINORVERSION)
3446	error = NFSERR_NOTSUPP;
3447	else if (i == 0)
3448	error = NFSERR_OPNOTINSESS;
3449	NFSCL_DEBUG(4, "off=%ju len=%ju sq=%u err=%d\n",
3450	(uintmax_t)off, (uintmax_t)len,
3451	stateid.seqid, error);
3452	if (error == 0) {
3453	NFSLOCKCLSTATE();
3454	clp = nfscl_getclntsess(sessionid);
3455	NFSCL_DEBUG(4, "cbly clp=%p\n", clp);
3456	if (clp != NULL) {
3457	lyp = nfscl_findlayout(clp,
3458	nfhp->nfh_fh,
3459	nfhp->nfh_len);
3460	NFSCL_DEBUG(4, "cblyp=%p\n",
3461	lyp);
3462	if (lyp != NULL &&
3463	(lyp->nfsly_flags &
3464	(NFSLY_FILES \|
3465	NFSLY_FLEXFILE)) != 0 &&
3466	!NFSBCMP(stateid.other,
3467	lyp->nfsly_stateid.other,
3468	NFSX_STATEIDOTHER)) {
3469	error =
3470	nfscl_layoutrecall(
3471	recalltype,
3472	lyp, iomode, off,
3473	len, stateid.seqid,
3474	0, 0, NULL,
3475	recallp);
3476	recallp = NULL;
3477	wakeup(clp);
3478	NFSCL_DEBUG(4,
3479	"aft layrcal=%d\n",
3480	error);
3481	} else
3482	error =
3483	NFSERR_NOMATCHLAYOUT;
3484	} else
3485	error = NFSERR_NOMATCHLAYOUT;
3486	NFSUNLOCKCLSTATE();
3487	}
3488	free(nfhp, M_NFSFH);
3489	} else if (recalltype == NFSLAYOUTRETURN_FSID) {
3490	NFSM_DISSECT(tl, uint32_t , 2 NFSX_HYPER);
3491	filesid[0] = fxdr_hyper(tl); tl += 2;
3492	filesid[1] = fxdr_hyper(tl); tl += 2;
3493	gotone = 0;
3494	NFSLOCKCLSTATE();
3495	clp = nfscl_getclntsess(sessionid);
3496	if (clp != NULL) {
3497	TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3498	nfsly_list) {
3499	if (lyp->nfsly_filesid[0] ==
3500	filesid[0] &&
3501	lyp->nfsly_filesid[1] ==
3502	filesid[1]) {
3503	error =
3504	nfscl_layoutrecall(
3505	recalltype,
3506	lyp, iomode, 0,
3507	UINT64_MAX,
3508	lyp->nfsly_stateid.seqid,
3509	0, 0, NULL,
3510	recallp);
3511	recallp = NULL;
3512	gotone = 1;
3513	}
3514	}
3515	if (gotone != 0)
3516	wakeup(clp);
3517	else
3518	error = NFSERR_NOMATCHLAYOUT;
3519	} else
3520	error = NFSERR_NOMATCHLAYOUT;
3521	NFSUNLOCKCLSTATE();
3522	} else if (recalltype == NFSLAYOUTRETURN_ALL) {
3523	gotone = 0;
3524	NFSLOCKCLSTATE();
3525	clp = nfscl_getclntsess(sessionid);
3526	if (clp != NULL) {
3527	TAILQ_FOREACH(lyp, &clp->nfsc_layout,
3528	nfsly_list) {
3529	error = nfscl_layoutrecall(
3530	recalltype, lyp, iomode, 0,
3531	UINT64_MAX,
3532	lyp->nfsly_stateid.seqid,
3533	0, 0, NULL, recallp);
3534	recallp = NULL;
3535	gotone = 1;
3536	}
3537	if (gotone != 0)
3538	wakeup(clp);
3539	else
3540	error = NFSERR_NOMATCHLAYOUT;
3541	} else
3542	error = NFSERR_NOMATCHLAYOUT;
3543	NFSUNLOCKCLSTATE();
3544	} else
3545	error = NFSERR_NOMATCHLAYOUT;
3546	if (recallp != NULL) {
3547	free(recallp, M_NFSLAYRECALL);
3548	recallp = NULL;
3549	}
3550	break;
3551	case NFSV4OP_CBSEQUENCE:
3552	NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3553	5 * NFSX_UNSIGNED);
3554	bcopy(tl, sessionid, NFSX_V4SESSIONID);
3555	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3556	seqid = fxdr_unsigned(uint32_t, *tl++);
3557	slotid = fxdr_unsigned(uint32_t, *tl++);
3558	highslot = fxdr_unsigned(uint32_t, *tl++);
3559	cachethis = *tl++;
3560	/* Throw away the referring call stuff. */
3561	clist = fxdr_unsigned(int, *tl);
3562	for (j = 0; j < clist; j++) {
3563	NFSM_DISSECT(tl, uint32_t *, NFSX_V4SESSIONID +
3564	NFSX_UNSIGNED);
3565	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3566	rcalls = fxdr_unsigned(int, *tl);
3567	for (k = 0; k < rcalls; k++) {
3568	NFSM_DISSECT(tl, uint32_t *,
3569	2 * NFSX_UNSIGNED);
3570	}
3571	}
3572	NFSLOCKCLSTATE();
3573	if (i == 0) {
3574	clp = nfscl_getclntsess(sessionid);
3575	if (clp == NULL)
3576	error = NFSERR_SERVERFAULT;
3577	} else
3578	error = NFSERR_SEQUENCEPOS;
3579	if (error == 0) {
3580	tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3581	error = nfsv4_seqsession(seqid, slotid,
3582	highslot, tsep->nfsess_cbslots, &rep,
3583	tsep->nfsess_backslots);
3584	}
3585	NFSUNLOCKCLSTATE();
3586	if (error == 0 \|\| error == NFSERR_REPLYFROMCACHE) {
3587	gotseq_ok = 1;
3588	if (rep != NULL) {
3589	/*
3590	* Handle a reply for a retried
3591	* callback. The reply will be
3592	* re-inserted in the session cache
3593	* by the nfsv4_seqsess_cacherep() call
3594	* after out:
3595	*/
3596	KASSERT(error == NFSERR_REPLYFROMCACHE,
3597	("cbsequence: non-NULL rep"));
3598	NFSCL_DEBUG(4, "Got cbretry\n");
3599	m_freem(nd->nd_mreq);
3600	nd->nd_mreq = rep;
3601	rep = NULL;
3602	goto out;
3603	}
3604	NFSM_BUILD(tl, uint32_t *,
3605	NFSX_V4SESSIONID + 4 * NFSX_UNSIGNED);
3606	bcopy(sessionid, tl, NFSX_V4SESSIONID);
3607	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
3608	*tl++ = txdr_unsigned(seqid);
3609	*tl++ = txdr_unsigned(slotid);
3610	*tl++ = txdr_unsigned(NFSV4_CBSLOTS - 1);
3611	*tl = txdr_unsigned(NFSV4_CBSLOTS - 1);
3612	}
3613	break;
3614	default:
3615	if (i == 0 && minorvers == NFSV41_MINORVERSION)
3616	error = NFSERR_OPNOTINSESS;
3617	else {
3618	NFSCL_DEBUG(1, "unsupp callback %d\n", op);
3619	error = NFSERR_NOTSUPP;
3620	}
3621	break;
3622	}
3623	if (error) {
3624	if (error == EBADRPC \|\| error == NFSERR_BADXDR) {
3625	nd->nd_repstat = NFSERR_BADXDR;
3626	} else {
3627	nd->nd_repstat = error;
3628	}
3629	error = 0;
3630	}
3631	retops++;
3632	if (nd->nd_repstat) {
3633	*repp = nfscl_errmap(nd, minorvers);
3634	break;
3635	} else
3636	repp = 0; / NFS4_OK */
3637	}
3638	nfsmout:
3639	if (recallp != NULL)
3640	free(recallp, M_NFSLAYRECALL);
3641	if (error) {
3642	if (error == EBADRPC \|\| error == NFSERR_BADXDR)
3643	nd->nd_repstat = NFSERR_BADXDR;
3644	else
3645	printf("nfsv4 comperr1=%d\n", error);
3646	}
3647	if (taglen == -1) {
3648	NFSM_BUILD(tl, u_int32_t , 2 NFSX_UNSIGNED);
3649	*tl++ = 0;
3650	*tl = 0;
3651	} else {
3652	*retopsp = txdr_unsigned(retops);
3653	}
3654	*nd->nd_errp = nfscl_errmap(nd, minorvers);
3655	out:
3656	if (gotseq_ok != 0) {
3657	rep = m_copym(nd->nd_mreq, 0, M_COPYALL, M_WAITOK);
3658	NFSLOCKCLSTATE();
3659	clp = nfscl_getclntsess(sessionid);
3660	if (clp != NULL) {
3661	tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3662	nfsv4_seqsess_cacherep(slotid, tsep->nfsess_cbslots,
3663	NFSERR_OK, &rep);
3664	NFSUNLOCKCLSTATE();
3665	} else {
3666	NFSUNLOCKCLSTATE();
3667	m_freem(rep);
3668	}
3669	}
3670	}
3671
3672	/*
3673	* Generate the next cbident value. Basically just increment a static value
3674	* and then check that it isn't already in the list, if it has wrapped around.
3675	*/
3676	static u_int32_t
3677	nfscl_nextcbident(void)
3678	{
3679	struct nfsclclient *clp;
3680	int matched;
3681	static u_int32_t nextcbident = 0;
3682	static int haswrapped = 0;
3683
3684	nextcbident++;
3685	if (nextcbident == 0)
3686	haswrapped = 1;
3687	if (haswrapped) {
3688	/*
3689	* Search the clientid list for one already using this cbident.
3690	*/
3691	do {
3692	matched = 0;
3693	NFSLOCKCLSTATE();
3694	LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3695	if (clp->nfsc_cbident == nextcbident) {
3696	matched = 1;
3697	break;
3698	}
3699	}
3700	NFSUNLOCKCLSTATE();
3701	if (matched == 1)
3702	nextcbident++;
3703	} while (matched);
3704	}
3705	return (nextcbident);
3706	}
3707
3708	/*
3709	* Get the mount point related to a given cbident or session and busy it.
3710	*/
3711	static mount_t
3712	nfscl_getmnt(int minorvers, uint8_t *sessionid, u_int32_t cbident,
3713	struct nfsclclient **clpp)
3714	{
3715	struct nfsclclient *clp;
3716	mount_t mp;
3717	int error;
3718	struct nfsclsession *tsep;
3719
3720	*clpp = NULL;
3721	NFSLOCKCLSTATE();
3722	LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3723	tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3724	if (minorvers == NFSV4_MINORVERSION) {
3725	if (clp->nfsc_cbident == cbident)
3726	break;
3727	} else if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3728	NFSX_V4SESSIONID))
3729	break;
3730	}
3731	if (clp == NULL) {
3732	NFSUNLOCKCLSTATE();
3733	return (NULL);
3734	}
3735	mp = clp->nfsc_nmp->nm_mountp;
3736	vfs_ref(mp);
3737	NFSUNLOCKCLSTATE();
3738	error = vfs_busy(mp, 0);
3739	vfs_rel(mp);
3740	if (error != 0)
3741	return (NULL);
3742	*clpp = clp;
3743	return (mp);
3744	}
3745
3746	/*
3747	* Get the clientid pointer related to a given cbident.
3748	*/
3749	static struct nfsclclient *
3750	nfscl_getclnt(u_int32_t cbident)
3751	{
3752	struct nfsclclient *clp;
3753
3754	LIST_FOREACH(clp, &nfsclhead, nfsc_list)
3755	if (clp->nfsc_cbident == cbident)
3756	break;
3757	return (clp);
3758	}
3759
3760	/*
3761	* Get the clientid pointer related to a given sessionid.
3762	*/
3763	static struct nfsclclient *
3764	nfscl_getclntsess(uint8_t *sessionid)
3765	{
3766	struct nfsclclient *clp;
3767	struct nfsclsession *tsep;
3768
3769	LIST_FOREACH(clp, &nfsclhead, nfsc_list) {
3770	tsep = nfsmnt_mdssession(clp->nfsc_nmp);
3771	if (!NFSBCMP(tsep->nfsess_sessionid, sessionid,
3772	NFSX_V4SESSIONID))
3773	break;
3774	}
3775	return (clp);
3776	}
3777
3778	/*
3779	* Search for a lock conflict locally on the client. A conflict occurs if
3780	* - not same owner and overlapping byte range and at least one of them is
3781	* a write lock or this is an unlock.
3782	*/
3783	static int
3784	nfscl_localconflict(struct nfsclclient clp, u_int8_t fhp, int fhlen,
3785	struct nfscllock nlop, u_int8_t own, struct nfscldeleg *dp,
3786	struct nfscllock **lopp)
3787	{
3788	struct nfsclowner *owp;
3789	struct nfsclopen *op;
3790	int ret;
3791
3792	if (dp != NULL) {
3793	ret = nfscl_checkconflict(&dp->nfsdl_lock, nlop, own, lopp);
3794	if (ret)
3795	return (ret);
3796	}
3797	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3798	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
3799	if (op->nfso_fhlen == fhlen &&
3800	!NFSBCMP(op->nfso_fh, fhp, fhlen)) {
3801	ret = nfscl_checkconflict(&op->nfso_lock, nlop,
3802	own, lopp);
3803	if (ret)
3804	return (ret);
3805	}
3806	}
3807	}
3808	return (0);
3809	}
3810
3811	static int
3812	nfscl_checkconflict(struct nfscllockownerhead lhp, struct nfscllock nlop,
3813	u_int8_t own, struct nfscllock *lopp)
3814	{
3815	struct nfscllockowner *lp;
3816	struct nfscllock *lop;
3817
3818	LIST_FOREACH(lp, lhp, nfsl_list) {
3819	if (NFSBCMP(lp->nfsl_owner, own, NFSV4CL_LOCKNAMELEN)) {
3820	LIST_FOREACH(lop, &lp->nfsl_lock, nfslo_list) {
3821	if (lop->nfslo_first >= nlop->nfslo_end)
3822	break;
3823	if (lop->nfslo_end <= nlop->nfslo_first)
3824	continue;
3825	if (lop->nfslo_type == F_WRLCK \|\|
3826	nlop->nfslo_type == F_WRLCK \|\|
3827	nlop->nfslo_type == F_UNLCK) {
3828	if (lopp != NULL)
3829	*lopp = lop;
3830	return (NFSERR_DENIED);
3831	}
3832	}
3833	}
3834	}
3835	return (0);
3836	}
3837
3838	/*
3839	* Check for a local conflicting lock.
3840	*/
3841	APPLESTATIC int
3842	nfscl_lockt(vnode_t vp, struct nfsclclient *clp, u_int64_t off,
3843	u_int64_t len, struct flock fl, NFSPROC_T p, void *id, int flags)
3844	{
3845	struct nfscllock *lop, nlck;
3846	struct nfscldeleg *dp;
3847	struct nfsnode *np;
3848	u_int8_t own[NFSV4CL_LOCKNAMELEN];
3849	int error;
3850
3851	nlck.nfslo_type = fl->l_type;
3852	nlck.nfslo_first = off;
3853	if (len == NFS64BITSSET) {
3854	nlck.nfslo_end = NFS64BITSSET;
3855	} else {
3856	nlck.nfslo_end = off + len;
3857	if (nlck.nfslo_end <= nlck.nfslo_first)
3858	return (NFSERR_INVAL);
3859	}
3860	np = VTONFS(vp);
3861	nfscl_filllockowner(id, own, flags);
3862	NFSLOCKCLSTATE();
3863	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
3864	error = nfscl_localconflict(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len,
3865	&nlck, own, dp, &lop);
3866	if (error != 0) {
3867	fl->l_whence = SEEK_SET;
3868	fl->l_start = lop->nfslo_first;
3869	if (lop->nfslo_end == NFS64BITSSET)
3870	fl->l_len = 0;
3871	else
3872	fl->l_len = lop->nfslo_end - lop->nfslo_first;
3873	fl->l_pid = (pid_t)0;
3874	fl->l_type = lop->nfslo_type;
3875	error = -1; /* no RPC required */
3876	} else if (dp != NULL && ((dp->nfsdl_flags & NFSCLDL_WRITE) \|\|
3877	fl->l_type == F_RDLCK)) {
3878	/*
3879	* The delegation ensures that there isn't a conflicting
3880	* lock on the server, so return -1 to indicate an RPC
3881	* isn't required.
3882	*/
3883	fl->l_type = F_UNLCK;
3884	error = -1;
3885	}
3886	NFSUNLOCKCLSTATE();
3887	return (error);
3888	}
3889
3890	/*
3891	* Handle Recall of a delegation.
3892	* The clp must be exclusive locked when this is called.
3893	*/
3894	static int
3895	nfscl_recalldeleg(struct nfsclclient clp, struct nfsmount nmp,
3896	struct nfscldeleg dp, vnode_t vp, struct ucred cred, NFSPROC_T *p,
3897	int called_from_renewthread)
3898	{
3899	struct nfsclowner owp, lowp, *nowp;
3900	struct nfsclopen op, lop;
3901	struct nfscllockowner *lp;
3902	struct nfscllock *lckp;
3903	struct nfsnode *np;
3904	int error = 0, ret, gotvp = 0;
3905
3906	if (vp == NULL) {
3907	/*
3908	* First, get a vnode for the file. This is needed to do RPCs.
3909	*/
3910	ret = nfscl_ngetreopen(nmp->nm_mountp, dp->nfsdl_fh,
3911	dp->nfsdl_fhlen, p, &np);
3912	if (ret) {
3913	/*
3914	* File isn't open, so nothing to move over to the
3915	* server.
3916	*/
3917	return (0);
3918	}
3919	vp = NFSTOV(np);
3920	gotvp = 1;
3921	} else {
3922	np = VTONFS(vp);
3923	}
3924	dp->nfsdl_flags &= ~NFSCLDL_MODTIMESET;
3925
3926	/*
3927	* Ok, if it's a write delegation, flush data to the server, so
3928	* that close/open consistency is retained.
3929	*/
3930	ret = 0;
3931	NFSLOCKNODE(np);
3932	if ((dp->nfsdl_flags & NFSCLDL_WRITE) && (np->n_flag & NMODIFIED)) {
3933	np->n_flag \|= NDELEGRECALL;
3934	NFSUNLOCKNODE(np);
3935	ret = ncl_flush(vp, MNT_WAIT, p, 1, called_from_renewthread);
3936	NFSLOCKNODE(np);
3937	np->n_flag &= ~NDELEGRECALL;
3938	}
3939	NFSINVALATTRCACHE(np);
3940	NFSUNLOCKNODE(np);
3941	if (ret == EIO && called_from_renewthread != 0) {
3942	/*
3943	* If the flush failed with EIO for the renew thread,
3944	* return now, so that the dirty buffer will be flushed
3945	* later.
3946	*/
3947	if (gotvp != 0)
3948	vrele(vp);
3949	return (ret);
3950	}
3951
3952	/*
3953	* Now, for each openowner with opens issued locally, move them
3954	* over to state against the server.
3955	*/
3956	LIST_FOREACH(lowp, &dp->nfsdl_owner, nfsow_list) {
3957	lop = LIST_FIRST(&lowp->nfsow_open);
3958	if (lop != NULL) {
3959	if (LIST_NEXT(lop, nfso_list) != NULL)
3960	panic("nfsdlg mult opens");
3961	/*
3962	* Look for the same openowner against the server.
3963	*/
3964	LIST_FOREACH(owp, &clp->nfsc_owner, nfsow_list) {
3965	if (!NFSBCMP(lowp->nfsow_owner,
3966	owp->nfsow_owner, NFSV4CL_LOCKNAMELEN)) {
3967	newnfs_copycred(&dp->nfsdl_cred, cred);
3968	ret = nfscl_moveopen(vp, clp, nmp, lop,
3969	owp, dp, cred, p);
3970	if (ret == NFSERR_STALECLIENTID \|\|
3971	ret == NFSERR_STALEDONTRECOVER \|\|
3972	ret == NFSERR_BADSESSION) {
3973	if (gotvp)
3974	vrele(vp);
3975	return (ret);
3976	}
3977	if (ret) {
3978	nfscl_freeopen(lop, 1);
3979	if (!error)
3980	error = ret;
3981	}
3982	break;
3983	}
3984	}
3985
3986	/*
3987	* If no openowner found, create one and get an open
3988	* for it.
3989	*/
3990	if (owp == NULL) {
3991	nowp = malloc(
3992	sizeof (struct nfsclowner), M_NFSCLOWNER,
3993	M_WAITOK);
3994	nfscl_newopen(clp, NULL, &owp, &nowp, &op,
3995	NULL, lowp->nfsow_owner, dp->nfsdl_fh,
3996	dp->nfsdl_fhlen, NULL, NULL);
3997	newnfs_copycred(&dp->nfsdl_cred, cred);
3998	ret = nfscl_moveopen(vp, clp, nmp, lop,
3999	owp, dp, cred, p);
4000	if (ret) {
4001	nfscl_freeopenowner(owp, 0);
4002	if (ret == NFSERR_STALECLIENTID \|\|
4003	ret == NFSERR_STALEDONTRECOVER \|\|
4004	ret == NFSERR_BADSESSION) {
4005	if (gotvp)
4006	vrele(vp);
4007	return (ret);
4008	}
4009	if (ret) {
4010	nfscl_freeopen(lop, 1);
4011	if (!error)
4012	error = ret;
4013	}
4014	}
4015	}
4016	}
4017	}
4018
4019	/*
4020	* Now, get byte range locks for any locks done locally.
4021	*/
4022	LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4023	LIST_FOREACH(lckp, &lp->nfsl_lock, nfslo_list) {
4024	newnfs_copycred(&dp->nfsdl_cred, cred);
4025	ret = nfscl_relock(vp, clp, nmp, lp, lckp, cred, p);
4026	if (ret == NFSERR_STALESTATEID \|\|
4027	ret == NFSERR_STALEDONTRECOVER \|\|
4028	ret == NFSERR_STALECLIENTID \|\|
4029	ret == NFSERR_BADSESSION) {
4030	if (gotvp)
4031	vrele(vp);
4032	return (ret);
4033	}
4034	if (ret && !error)
4035	error = ret;
4036	}
4037	}
4038	if (gotvp)
4039	vrele(vp);
4040	return (error);
4041	}
4042
4043	/*
4044	* Move a locally issued open over to an owner on the state list.
4045	* SIDE EFFECT: If it needs to sleep (do an rpc), it unlocks clstate and
4046	* returns with it unlocked.
4047	*/
4048	static int
4049	nfscl_moveopen(vnode_t vp, struct nfsclclient clp, struct nfsmount nmp,
4050	struct nfsclopen lop, struct nfsclowner owp, struct nfscldeleg *dp,
4051	struct ucred cred, NFSPROC_T p)
4052	{
4053	struct nfsclopen op, nop;
4054	struct nfscldeleg *ndp;
4055	struct nfsnode *np;
4056	int error = 0, newone;
4057
4058	/*
4059	* First, look for an appropriate open, If found, just increment the
4060	* opencnt in it.
4061	*/
4062	LIST_FOREACH(op, &owp->nfsow_open, nfso_list) {
4063	if ((op->nfso_mode & lop->nfso_mode) == lop->nfso_mode &&
4064	op->nfso_fhlen == lop->nfso_fhlen &&
4065	!NFSBCMP(op->nfso_fh, lop->nfso_fh, op->nfso_fhlen)) {
4066	op->nfso_opencnt += lop->nfso_opencnt;
4067	nfscl_freeopen(lop, 1);
4068	return (0);
4069	}
4070	}
4071
4072	/* No appropriate open, so we have to do one against the server. */
4073	np = VTONFS(vp);
4074	nop = malloc(sizeof (struct nfsclopen) +
4075	lop->nfso_fhlen - 1, M_NFSCLOPEN, M_WAITOK);
4076	newone = 0;
4077	nfscl_newopen(clp, NULL, &owp, NULL, &op, &nop, owp->nfsow_owner,
4078	lop->nfso_fh, lop->nfso_fhlen, cred, &newone);
4079	ndp = dp;
4080	error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data, np->n_v4->n4_fhlen,
4081	lop->nfso_fh, lop->nfso_fhlen, lop->nfso_mode, op,
4082	NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, &ndp, 0, 0, cred, p);
4083	if (error) {
4084	if (newone)
4085	nfscl_freeopen(op, 0);
4086	} else {
4087	op->nfso_mode \|= lop->nfso_mode;
4088	op->nfso_opencnt += lop->nfso_opencnt;
4089	nfscl_freeopen(lop, 1);
4090	}
4091	if (nop != NULL)
4092	free(nop, M_NFSCLOPEN);
4093	if (ndp != NULL) {
4094	/*
4095	* What should I do with the returned delegation, since the
4096	* delegation is being recalled? For now, just printf and
4097	* through it away.
4098	*/
4099	printf("Moveopen returned deleg\n");
4100	free(ndp, M_NFSCLDELEG);
4101	}
4102	return (error);
4103	}
4104
4105	/*
4106	* Recall all delegations on this client.
4107	*/
4108	static void
4109	nfscl_totalrecall(struct nfsclclient *clp)
4110	{
4111	struct nfscldeleg *dp;
4112
4113	TAILQ_FOREACH(dp, &clp->nfsc_deleg, nfsdl_list) {
4114	if ((dp->nfsdl_flags & NFSCLDL_DELEGRET) == 0)
4115	dp->nfsdl_flags \|= NFSCLDL_RECALL;
4116	}
4117	}
4118
4119	/*
4120	* Relock byte ranges. Called for delegation recall and state expiry.
4121	*/
4122	static int
4123	nfscl_relock(vnode_t vp, struct nfsclclient clp, struct nfsmount nmp,
4124	struct nfscllockowner lp, struct nfscllock lop, struct ucred *cred,
4125	NFSPROC_T *p)
4126	{
4127	struct nfscllockowner *nlp;
4128	struct nfsfh *nfhp;
4129	u_int64_t off, len;
4130	int error, newone, donelocally;
4131
4132	off = lop->nfslo_first;
4133	len = lop->nfslo_end - lop->nfslo_first;
4134	error = nfscl_getbytelock(vp, off, len, lop->nfslo_type, cred, p,
4135	clp, 1, NULL, lp->nfsl_lockflags, lp->nfsl_owner,
4136	lp->nfsl_openowner, &nlp, &newone, &donelocally);
4137	if (error \|\| donelocally)
4138	return (error);
4139	nfhp = VTONFS(vp)->n_fhp;
4140	error = nfscl_trylock(nmp, vp, nfhp->nfh_fh,
4141	nfhp->nfh_len, nlp, newone, 0, off,
4142	len, lop->nfslo_type, cred, p);
4143	if (error)
4144	nfscl_freelockowner(nlp, 0);
4145	return (error);
4146	}
4147
4148	/*
4149	* Called to re-open a file. Basically get a vnode for the file handle
4150	* and then call nfsrpc_openrpc() to do the rest.
4151	*/
4152	static int
4153	nfsrpc_reopen(struct nfsmount nmp, u_int8_t fhp, int fhlen,
4154	u_int32_t mode, struct nfsclopen op, struct nfscldeleg *dpp,
4155	struct ucred cred, NFSPROC_T p)
4156	{
4157	struct nfsnode *np;
4158	vnode_t vp;
4159	int error;
4160
4161	error = nfscl_ngetreopen(nmp->nm_mountp, fhp, fhlen, p, &np);
4162	if (error)
4163	return (error);
4164	vp = NFSTOV(np);
4165	if (np->n_v4 != NULL) {
4166	error = nfscl_tryopen(nmp, vp, np->n_v4->n4_data,
4167	np->n_v4->n4_fhlen, fhp, fhlen, mode, op,
4168	NFS4NODENAME(np->n_v4), np->n_v4->n4_namelen, dpp, 0, 0,
4169	cred, p);
4170	} else {
4171	error = EINVAL;
4172	}
4173	vrele(vp);
4174	return (error);
4175	}
4176
4177	/*
4178	* Try an open against the server. Just call nfsrpc_openrpc(), retrying while
4179	* NFSERR_DELAY. Also, try system credentials, if the passed in credentials
4180	* fail.
4181	*/
4182	static int
4183	nfscl_tryopen(struct nfsmount nmp, vnode_t vp, u_int8_t fhp, int fhlen,
4184	u_int8_t newfhp, int newfhlen, u_int32_t mode, struct nfsclopen op,
4185	u_int8_t name, int namelen, struct nfscldeleg *ndpp,
4186	int reclaim, u_int32_t delegtype, struct ucred cred, NFSPROC_T p)
4187	{
4188	int error;
4189
4190	do {
4191	error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp, newfhlen,
4192	mode, op, name, namelen, ndpp, reclaim, delegtype, cred, p,
4193	0, 0);
4194	if (error == NFSERR_DELAY)
4195	(void) nfs_catnap(PZERO, error, "nfstryop");
4196	} while (error == NFSERR_DELAY);
4197	if (error == EAUTH \|\| error == EACCES) {
4198	/* Try again using system credentials */
4199	newnfs_setroot(cred);
4200	do {
4201	error = nfsrpc_openrpc(nmp, vp, fhp, fhlen, newfhp,
4202	newfhlen, mode, op, name, namelen, ndpp, reclaim,
4203	delegtype, cred, p, 1, 0);
4204	if (error == NFSERR_DELAY)
4205	(void) nfs_catnap(PZERO, error, "nfstryop");
4206	} while (error == NFSERR_DELAY);
4207	}
4208	return (error);
4209	}
4210
4211	/*
4212	* Try a byte range lock. Just loop on nfsrpc_lock() while it returns
4213	* NFSERR_DELAY. Also, retry with system credentials, if the provided
4214	* cred don't work.
4215	*/
4216	static int
4217	nfscl_trylock(struct nfsmount nmp, vnode_t vp, u_int8_t fhp,
4218	int fhlen, struct nfscllockowner *nlp, int newone, int reclaim,
4219	u_int64_t off, u_int64_t len, short type, struct ucred cred, NFSPROC_T p)
4220	{
4221	struct nfsrv_descript nfsd, *nd = &nfsd;
4222	int error;
4223
4224	do {
4225	error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp, newone,
4226	reclaim, off, len, type, cred, p, 0);
4227	if (!error && nd->nd_repstat == NFSERR_DELAY)
4228	(void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4229	"nfstrylck");
4230	} while (!error && nd->nd_repstat == NFSERR_DELAY);
4231	if (!error)
4232	error = nd->nd_repstat;
4233	if (error == EAUTH \|\| error == EACCES) {
4234	/* Try again using root credentials */
4235	newnfs_setroot(cred);
4236	do {
4237	error = nfsrpc_lock(nd, nmp, vp, fhp, fhlen, nlp,
4238	newone, reclaim, off, len, type, cred, p, 1);
4239	if (!error && nd->nd_repstat == NFSERR_DELAY)
4240	(void) nfs_catnap(PZERO, (int)nd->nd_repstat,
4241	"nfstrylck");
4242	} while (!error && nd->nd_repstat == NFSERR_DELAY);
4243	if (!error)
4244	error = nd->nd_repstat;
4245	}
4246	return (error);
4247	}
4248
4249	/*
4250	* Try a delegreturn against the server. Just call nfsrpc_delegreturn(),
4251	* retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4252	* credentials fail.
4253	*/
4254	static int
4255	nfscl_trydelegreturn(struct nfscldeleg dp, struct ucred cred,
4256	struct nfsmount nmp, NFSPROC_T p)
4257	{
4258	int error;
4259
4260	do {
4261	error = nfsrpc_delegreturn(dp, cred, nmp, p, 0);
4262	if (error == NFSERR_DELAY)
4263	(void) nfs_catnap(PZERO, error, "nfstrydp");
4264	} while (error == NFSERR_DELAY);
4265	if (error == EAUTH \|\| error == EACCES) {
4266	/* Try again using system credentials */
4267	newnfs_setroot(cred);
4268	do {
4269	error = nfsrpc_delegreturn(dp, cred, nmp, p, 1);
4270	if (error == NFSERR_DELAY)
4271	(void) nfs_catnap(PZERO, error, "nfstrydp");
4272	} while (error == NFSERR_DELAY);
4273	}
4274	return (error);
4275	}
4276
4277	/*
4278	* Try a close against the server. Just call nfsrpc_closerpc(),
4279	* retrying while NFSERR_DELAY. Also, try system credentials, if the passed in
4280	* credentials fail.
4281	*/
4282	APPLESTATIC int
4283	nfscl_tryclose(struct nfsclopen op, struct ucred cred,
4284	struct nfsmount nmp, NFSPROC_T p)
4285	{
4286	struct nfsrv_descript nfsd, *nd = &nfsd;
4287	int error;
4288
4289	do {
4290	error = nfsrpc_closerpc(nd, nmp, op, cred, p, 0);
4291	if (error == NFSERR_DELAY)
4292	(void) nfs_catnap(PZERO, error, "nfstrycl");
4293	} while (error == NFSERR_DELAY);
4294	if (error == EAUTH \|\| error == EACCES) {
4295	/* Try again using system credentials */
4296	newnfs_setroot(cred);
4297	do {
4298	error = nfsrpc_closerpc(nd, nmp, op, cred, p, 1);
4299	if (error == NFSERR_DELAY)
4300	(void) nfs_catnap(PZERO, error, "nfstrycl");
4301	} while (error == NFSERR_DELAY);
4302	}
4303	return (error);
4304	}
4305
4306	/*
4307	* Decide if a delegation on a file permits close without flushing writes
4308	* to the server. This might be a big performance win in some environments.
4309	* (Not useful until the client does caching on local stable storage.)
4310	*/
4311	APPLESTATIC int
4312	nfscl_mustflush(vnode_t vp)
4313	{
4314	struct nfsclclient *clp;
4315	struct nfscldeleg *dp;
4316	struct nfsnode *np;
4317	struct nfsmount *nmp;
4318
4319	np = VTONFS(vp);
4320	nmp = VFSTONFS(vnode_mount(vp));
4321	if (!NFSHASNFSV4(nmp))
4322	return (1);
4323	NFSLOCKCLSTATE();
4324	clp = nfscl_findcl(nmp);
4325	if (clp == NULL) {
4326	NFSUNLOCKCLSTATE();
4327	return (1);
4328	}
4329	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4330	if (dp != NULL && (dp->nfsdl_flags &
4331	(NFSCLDL_WRITE \| NFSCLDL_RECALL \| NFSCLDL_DELEGRET)) ==
4332	NFSCLDL_WRITE &&
4333	(dp->nfsdl_sizelimit >= np->n_size \|\|
4334	!NFSHASSTRICT3530(nmp))) {
4335	NFSUNLOCKCLSTATE();
4336	return (0);
4337	}
4338	NFSUNLOCKCLSTATE();
4339	return (1);
4340	}
4341
4342	/*
4343	* See if a (write) delegation exists for this file.
4344	*/
4345	APPLESTATIC int
4346	nfscl_nodeleg(vnode_t vp, int writedeleg)
4347	{
4348	struct nfsclclient *clp;
4349	struct nfscldeleg *dp;
4350	struct nfsnode *np;
4351	struct nfsmount *nmp;
4352
4353	np = VTONFS(vp);
4354	nmp = VFSTONFS(vnode_mount(vp));
4355	if (!NFSHASNFSV4(nmp))
4356	return (1);
4357	NFSLOCKCLSTATE();
4358	clp = nfscl_findcl(nmp);
4359	if (clp == NULL) {
4360	NFSUNLOCKCLSTATE();
4361	return (1);
4362	}
4363	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4364	if (dp != NULL &&
4365	(dp->nfsdl_flags & (NFSCLDL_RECALL \| NFSCLDL_DELEGRET)) == 0 &&
4366	(writedeleg == 0 \|\| (dp->nfsdl_flags & NFSCLDL_WRITE) ==
4367	NFSCLDL_WRITE)) {
4368	NFSUNLOCKCLSTATE();
4369	return (0);
4370	}
4371	NFSUNLOCKCLSTATE();
4372	return (1);
4373	}
4374
4375	/*
4376	* Look for an associated delegation that should be DelegReturned.
4377	*/
4378	APPLESTATIC int
4379	nfscl_removedeleg(vnode_t vp, NFSPROC_T p, nfsv4stateid_t stp)
4380	{
4381	struct nfsclclient *clp;
4382	struct nfscldeleg *dp;
4383	struct nfsclowner *owp;
4384	struct nfscllockowner *lp;
4385	struct nfsmount *nmp;
4386	struct ucred *cred;
4387	struct nfsnode *np;
4388	int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4389
4390	nmp = VFSTONFS(vnode_mount(vp));
4391	np = VTONFS(vp);
4392	NFSLOCKCLSTATE();
4393	/*
4394	* Loop around waiting for:
4395	* - outstanding I/O operations on delegations to complete
4396	* - for a delegation on vp that has state, lock the client and
4397	* do a recall
4398	* - return delegation with no state
4399	*/
4400	while (1) {
4401	clp = nfscl_findcl(nmp);
4402	if (clp == NULL) {
4403	NFSUNLOCKCLSTATE();
4404	return (retcnt);
4405	}
4406	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4407	np->n_fhp->nfh_len);
4408	if (dp != NULL) {
4409	/*
4410	* Wait for outstanding I/O ops to be done.
4411	*/
4412	if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4413	if (igotlock) {
4414	nfsv4_unlock(&clp->nfsc_lock, 0);
4415	igotlock = 0;
4416	}
4417	dp->nfsdl_rwlock.nfslock_lock \|= NFSV4LOCK_WANTED;
4418	(void) nfsmsleep(&dp->nfsdl_rwlock,
4419	NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4420	continue;
4421	}
4422	needsrecall = 0;
4423	LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4424	if (!LIST_EMPTY(&owp->nfsow_open)) {
4425	needsrecall = 1;
4426	break;
4427	}
4428	}
4429	if (!needsrecall) {
4430	LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4431	if (!LIST_EMPTY(&lp->nfsl_lock)) {
4432	needsrecall = 1;
4433	break;
4434	}
4435	}
4436	}
4437	if (needsrecall && !triedrecall) {
4438	dp->nfsdl_flags \|= NFSCLDL_DELEGRET;
4439	islept = 0;
4440	while (!igotlock) {
4441	igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4442	&islept, NFSCLSTATEMUTEXPTR, NULL);
4443	if (islept)
4444	break;
4445	}
4446	if (islept)
4447	continue;
4448	NFSUNLOCKCLSTATE();
4449	cred = newnfs_getcred();
4450	newnfs_copycred(&dp->nfsdl_cred, cred);
4451	(void) nfscl_recalldeleg(clp, nmp, dp, vp, cred, p, 0);
4452	NFSFREECRED(cred);
4453	triedrecall = 1;
4454	NFSLOCKCLSTATE();
4455	nfsv4_unlock(&clp->nfsc_lock, 0);
4456	igotlock = 0;
4457	continue;
4458	}
4459	*stp = dp->nfsdl_stateid;
4460	retcnt = 1;
4461	nfscl_cleandeleg(dp);
4462	nfscl_freedeleg(&clp->nfsc_deleg, dp);
4463	}
4464	if (igotlock)
4465	nfsv4_unlock(&clp->nfsc_lock, 0);
4466	NFSUNLOCKCLSTATE();
4467	return (retcnt);
4468	}
4469	}
4470
4471	/*
4472	* Look for associated delegation(s) that should be DelegReturned.
4473	*/
4474	APPLESTATIC int
4475	nfscl_renamedeleg(vnode_t fvp, nfsv4stateid_t fstp, int gotfdp, vnode_t tvp,
4476	nfsv4stateid_t tstp, int gottdp, NFSPROC_T *p)
4477	{
4478	struct nfsclclient *clp;
4479	struct nfscldeleg *dp;
4480	struct nfsclowner *owp;
4481	struct nfscllockowner *lp;
4482	struct nfsmount *nmp;
4483	struct ucred *cred;
4484	struct nfsnode *np;
4485	int igotlock = 0, triedrecall = 0, needsrecall, retcnt = 0, islept;
4486
4487	nmp = VFSTONFS(vnode_mount(fvp));
4488	*gotfdp = 0;
4489	*gottdp = 0;
4490	NFSLOCKCLSTATE();
4491	/*
4492	* Loop around waiting for:
4493	* - outstanding I/O operations on delegations to complete
4494	* - for a delegation on fvp that has state, lock the client and
4495	* do a recall
4496	* - return delegation(s) with no state.
4497	*/
4498	while (1) {
4499	clp = nfscl_findcl(nmp);
4500	if (clp == NULL) {
4501	NFSUNLOCKCLSTATE();
4502	return (retcnt);
4503	}
4504	np = VTONFS(fvp);
4505	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4506	np->n_fhp->nfh_len);
4507	if (dp != NULL && *gotfdp == 0) {
4508	/*
4509	* Wait for outstanding I/O ops to be done.
4510	*/
4511	if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4512	if (igotlock) {
4513	nfsv4_unlock(&clp->nfsc_lock, 0);
4514	igotlock = 0;
4515	}
4516	dp->nfsdl_rwlock.nfslock_lock \|= NFSV4LOCK_WANTED;
4517	(void) nfsmsleep(&dp->nfsdl_rwlock,
4518	NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4519	continue;
4520	}
4521	needsrecall = 0;
4522	LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4523	if (!LIST_EMPTY(&owp->nfsow_open)) {
4524	needsrecall = 1;
4525	break;
4526	}
4527	}
4528	if (!needsrecall) {
4529	LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4530	if (!LIST_EMPTY(&lp->nfsl_lock)) {
4531	needsrecall = 1;
4532	break;
4533	}
4534	}
4535	}
4536	if (needsrecall && !triedrecall) {
4537	dp->nfsdl_flags \|= NFSCLDL_DELEGRET;
4538	islept = 0;
4539	while (!igotlock) {
4540	igotlock = nfsv4_lock(&clp->nfsc_lock, 1,
4541	&islept, NFSCLSTATEMUTEXPTR, NULL);
4542	if (islept)
4543	break;
4544	}
4545	if (islept)
4546	continue;
4547	NFSUNLOCKCLSTATE();
4548	cred = newnfs_getcred();
4549	newnfs_copycred(&dp->nfsdl_cred, cred);
4550	(void) nfscl_recalldeleg(clp, nmp, dp, fvp, cred, p, 0);
4551	NFSFREECRED(cred);
4552	triedrecall = 1;
4553	NFSLOCKCLSTATE();
4554	nfsv4_unlock(&clp->nfsc_lock, 0);
4555	igotlock = 0;
4556	continue;
4557	}
4558	*fstp = dp->nfsdl_stateid;
4559	retcnt++;
4560	*gotfdp = 1;
4561	nfscl_cleandeleg(dp);
4562	nfscl_freedeleg(&clp->nfsc_deleg, dp);
4563	}
4564	if (igotlock) {
4565	nfsv4_unlock(&clp->nfsc_lock, 0);
4566	igotlock = 0;
4567	}
4568	if (tvp != NULL) {
4569	np = VTONFS(tvp);
4570	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh,
4571	np->n_fhp->nfh_len);
4572	if (dp != NULL && *gottdp == 0) {
4573	/*
4574	* Wait for outstanding I/O ops to be done.
4575	*/
4576	if (dp->nfsdl_rwlock.nfslock_usecnt > 0) {
4577	dp->nfsdl_rwlock.nfslock_lock \|= NFSV4LOCK_WANTED;
4578	(void) nfsmsleep(&dp->nfsdl_rwlock,
4579	NFSCLSTATEMUTEXPTR, PZERO, "nfscld", NULL);
4580	continue;
4581	}
4582	LIST_FOREACH(owp, &dp->nfsdl_owner, nfsow_list) {
4583	if (!LIST_EMPTY(&owp->nfsow_open)) {
4584	NFSUNLOCKCLSTATE();
4585	return (retcnt);
4586	}
4587	}
4588	LIST_FOREACH(lp, &dp->nfsdl_lock, nfsl_list) {
4589	if (!LIST_EMPTY(&lp->nfsl_lock)) {
4590	NFSUNLOCKCLSTATE();
4591	return (retcnt);
4592	}
4593	}
4594	*tstp = dp->nfsdl_stateid;
4595	retcnt++;
4596	*gottdp = 1;
4597	nfscl_cleandeleg(dp);
4598	nfscl_freedeleg(&clp->nfsc_deleg, dp);
4599	}
4600	}
4601	NFSUNLOCKCLSTATE();
4602	return (retcnt);
4603	}
4604	}
4605
4606	/*
4607	* Get a reference on the clientid associated with the mount point.
4608	* Return 1 if success, 0 otherwise.
4609	*/
4610	APPLESTATIC int
4611	nfscl_getref(struct nfsmount *nmp)
4612	{
4613	struct nfsclclient *clp;
4614
4615	NFSLOCKCLSTATE();
4616	clp = nfscl_findcl(nmp);
4617	if (clp == NULL) {
4618	NFSUNLOCKCLSTATE();
4619	return (0);
4620	}
4621	nfsv4_getref(&clp->nfsc_lock, NULL, NFSCLSTATEMUTEXPTR, NULL);
4622	NFSUNLOCKCLSTATE();
4623	return (1);
4624	}
4625
4626	/*
4627	* Release a reference on a clientid acquired with the above call.
4628	*/
4629	APPLESTATIC void
4630	nfscl_relref(struct nfsmount *nmp)
4631	{
4632	struct nfsclclient *clp;
4633
4634	NFSLOCKCLSTATE();
4635	clp = nfscl_findcl(nmp);
4636	if (clp == NULL) {
4637	NFSUNLOCKCLSTATE();
4638	return;
4639	}
4640	nfsv4_relref(&clp->nfsc_lock);
4641	NFSUNLOCKCLSTATE();
4642	}
4643
4644	/*
4645	* Save the size attribute in the delegation, since the nfsnode
4646	* is going away.
4647	*/
4648	APPLESTATIC void
4649	nfscl_reclaimnode(vnode_t vp)
4650	{
4651	struct nfsclclient *clp;
4652	struct nfscldeleg *dp;
4653	struct nfsnode *np = VTONFS(vp);
4654	struct nfsmount *nmp;
4655
4656	nmp = VFSTONFS(vnode_mount(vp));
4657	if (!NFSHASNFSV4(nmp))
4658	return;
4659	NFSLOCKCLSTATE();
4660	clp = nfscl_findcl(nmp);
4661	if (clp == NULL) {
4662	NFSUNLOCKCLSTATE();
4663	return;
4664	}
4665	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4666	if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4667	dp->nfsdl_size = np->n_size;
4668	NFSUNLOCKCLSTATE();
4669	}
4670
4671	/*
4672	* Get the saved size attribute in the delegation, since it is a
4673	* newly allocated nfsnode.
4674	*/
4675	APPLESTATIC void
4676	nfscl_newnode(vnode_t vp)
4677	{
4678	struct nfsclclient *clp;
4679	struct nfscldeleg *dp;
4680	struct nfsnode *np = VTONFS(vp);
4681	struct nfsmount *nmp;
4682
4683	nmp = VFSTONFS(vnode_mount(vp));
4684	if (!NFSHASNFSV4(nmp))
4685	return;
4686	NFSLOCKCLSTATE();
4687	clp = nfscl_findcl(nmp);
4688	if (clp == NULL) {
4689	NFSUNLOCKCLSTATE();
4690	return;
4691	}
4692	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4693	if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE))
4694	np->n_size = dp->nfsdl_size;
4695	NFSUNLOCKCLSTATE();
4696	}
4697
4698	/*
4699	* If there is a valid write delegation for this file, set the modtime
4700	* to the local clock time.
4701	*/
4702	APPLESTATIC void
4703	nfscl_delegmodtime(vnode_t vp)
4704	{
4705	struct nfsclclient *clp;
4706	struct nfscldeleg *dp;
4707	struct nfsnode *np = VTONFS(vp);
4708	struct nfsmount *nmp;
4709
4710	nmp = VFSTONFS(vnode_mount(vp));
4711	if (!NFSHASNFSV4(nmp))
4712	return;
4713	NFSLOCKCLSTATE();
4714	clp = nfscl_findcl(nmp);
4715	if (clp == NULL) {
4716	NFSUNLOCKCLSTATE();
4717	return;
4718	}
4719	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4720	if (dp != NULL && (dp->nfsdl_flags & NFSCLDL_WRITE)) {
4721	nanotime(&dp->nfsdl_modtime);
4722	dp->nfsdl_flags \|= NFSCLDL_MODTIMESET;
4723	}
4724	NFSUNLOCKCLSTATE();
4725	}
4726
4727	/*
4728	* If there is a valid write delegation for this file with a modtime set,
4729	* put that modtime in mtime.
4730	*/
4731	APPLESTATIC void
4732	nfscl_deleggetmodtime(vnode_t vp, struct timespec *mtime)
4733	{
4734	struct nfsclclient *clp;
4735	struct nfscldeleg *dp;
4736	struct nfsnode *np = VTONFS(vp);
4737	struct nfsmount *nmp;
4738
4739	nmp = VFSTONFS(vnode_mount(vp));
4740	if (!NFSHASNFSV4(nmp))
4741	return;
4742	NFSLOCKCLSTATE();
4743	clp = nfscl_findcl(nmp);
4744	if (clp == NULL) {
4745	NFSUNLOCKCLSTATE();
4746	return;
4747	}
4748	dp = nfscl_finddeleg(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
4749	if (dp != NULL &&
4750	(dp->nfsdl_flags & (NFSCLDL_WRITE \| NFSCLDL_MODTIMESET)) ==
4751	(NFSCLDL_WRITE \| NFSCLDL_MODTIMESET))
4752	*mtime = dp->nfsdl_modtime;
4753	NFSUNLOCKCLSTATE();
4754	}
4755
4756	static int
4757	nfscl_errmap(struct nfsrv_descript *nd, u_int32_t minorvers)
4758	{
4759	short defaulterrp, errp;
4760
4761	if (!nd->nd_repstat)
4762	return (0);
4763	if (nd->nd_procnum == NFSPROC_NOOP)
4764	return (txdr_unsigned(nd->nd_repstat & 0xffff));
4765	if (nd->nd_repstat == EBADRPC)
4766	return (txdr_unsigned(NFSERR_BADXDR));
4767	if (nd->nd_repstat == NFSERR_MINORVERMISMATCH \|\|
4768	nd->nd_repstat == NFSERR_OPILLEGAL)
4769	return (txdr_unsigned(nd->nd_repstat));
4770	if (nd->nd_repstat >= NFSERR_BADIOMODE && nd->nd_repstat < 20000 &&
4771	minorvers > NFSV4_MINORVERSION) {
4772	/* NFSv4.n error. */
4773	return (txdr_unsigned(nd->nd_repstat));
4774	}
4775	if (nd->nd_procnum < NFSV4OP_CBNOPS)
4776	errp = defaulterrp = nfscl_cberrmap[nd->nd_procnum];
4777	else
4778	return (txdr_unsigned(nd->nd_repstat));
4779	while (*++errp)
4780	if (*errp == (short)nd->nd_repstat)
4781	return (txdr_unsigned(nd->nd_repstat));
4782	return (txdr_unsigned(*defaulterrp));
4783	}
4784
4785	/*
4786	* Called to find/add a layout to a client.
4787	* This function returns the layout with a refcnt (shared lock) upon
4788	* success (returns 0) or with no lock/refcnt on the layout when an
4789	* error is returned.
4790	* If a layout is passed in via lypp, it is locked (exclusively locked).
4791	*/
4792	APPLESTATIC int
4793	nfscl_layout(struct nfsmount nmp, vnode_t vp, u_int8_t fhp, int fhlen,
4794	nfsv4stateid_t *stateidp, int layouttype, int retonclose,
4795	struct nfsclflayouthead fhlp, struct nfscllayout *lypp,
4796	struct ucred cred, NFSPROC_T p)
4797	{
4798	struct nfsclclient *clp;
4799	struct nfscllayout lyp, tlyp;
4800	struct nfsclflayout *flp;
4801	struct nfsnode *np = VTONFS(vp);
4802	mount_t mp;
4803	int layout_passed_in;
4804
4805	mp = nmp->nm_mountp;
4806	layout_passed_in = 1;
4807	tlyp = NULL;
4808	lyp = *lypp;
4809	if (lyp == NULL) {
4810	layout_passed_in = 0;
4811	tlyp = malloc(sizeof(*tlyp) + fhlen - 1, M_NFSLAYOUT,
4812	M_WAITOK \| M_ZERO);
4813	}
4814
4815	NFSLOCKCLSTATE();
4816	clp = nmp->nm_clp;
4817	if (clp == NULL) {
4818	if (layout_passed_in != 0)
4819	nfsv4_unlock(&lyp->nfsly_lock, 0);
4820	NFSUNLOCKCLSTATE();
4821	if (tlyp != NULL)
4822	free(tlyp, M_NFSLAYOUT);
4823	return (EPERM);
4824	}
4825	if (lyp == NULL) {
4826	/*
4827	* Although no lyp was passed in, another thread might have
4828	* allocated one. If one is found, just increment it's ref
4829	* count and return it.
4830	*/
4831	lyp = nfscl_findlayout(clp, fhp, fhlen);
4832	if (lyp == NULL) {
4833	lyp = tlyp;
4834	tlyp = NULL;
4835	lyp->nfsly_stateid.seqid = stateidp->seqid;
4836	lyp->nfsly_stateid.other[0] = stateidp->other[0];
4837	lyp->nfsly_stateid.other[1] = stateidp->other[1];
4838	lyp->nfsly_stateid.other[2] = stateidp->other[2];
4839	lyp->nfsly_lastbyte = 0;
4840	LIST_INIT(&lyp->nfsly_flayread);
4841	LIST_INIT(&lyp->nfsly_flayrw);
4842	LIST_INIT(&lyp->nfsly_recall);
4843	lyp->nfsly_filesid[0] = np->n_vattr.na_filesid[0];
4844	lyp->nfsly_filesid[1] = np->n_vattr.na_filesid[1];
4845	lyp->nfsly_clp = clp;
4846	if (layouttype == NFSLAYOUT_FLEXFILE)
4847	lyp->nfsly_flags = NFSLY_FLEXFILE;
4848	else
4849	lyp->nfsly_flags = NFSLY_FILES;
4850	if (retonclose != 0)
4851	lyp->nfsly_flags \|= NFSLY_RETONCLOSE;
4852	lyp->nfsly_fhlen = fhlen;
4853	NFSBCOPY(fhp, lyp->nfsly_fh, fhlen);
4854	TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4855	LIST_INSERT_HEAD(NFSCLLAYOUTHASH(clp, fhp, fhlen), lyp,
4856	nfsly_hash);
4857	lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4858	nfscl_layoutcnt++;
4859	} else {
4860	if (retonclose != 0)
4861	lyp->nfsly_flags \|= NFSLY_RETONCLOSE;
4862	TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4863	TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4864	lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4865	}
4866	nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
4867	if (NFSCL_FORCEDISM(mp)) {
4868	NFSUNLOCKCLSTATE();
4869	if (tlyp != NULL)
4870	free(tlyp, M_NFSLAYOUT);
4871	return (EPERM);
4872	}
4873	*lypp = lyp;
4874	} else
4875	lyp->nfsly_stateid.seqid = stateidp->seqid;
4876
4877	/* Merge the new list of File Layouts into the list. */
4878	flp = LIST_FIRST(fhlp);
4879	if (flp != NULL) {
4880	if (flp->nfsfl_iomode == NFSLAYOUTIOMODE_READ)
4881	nfscl_mergeflayouts(&lyp->nfsly_flayread, fhlp);
4882	else
4883	nfscl_mergeflayouts(&lyp->nfsly_flayrw, fhlp);
4884	}
4885	if (layout_passed_in != 0)
4886	nfsv4_unlock(&lyp->nfsly_lock, 1);
4887	NFSUNLOCKCLSTATE();
4888	if (tlyp != NULL)
4889	free(tlyp, M_NFSLAYOUT);
4890	return (0);
4891	}
4892
4893	/*
4894	* Search for a layout by MDS file handle.
4895	* If one is found, it is returned with a refcnt (shared lock) iff
4896	* retflpp returned non-NULL and locked (exclusive locked) iff retflpp is
4897	* returned NULL.
4898	*/
4899	struct nfscllayout *
4900	nfscl_getlayout(struct nfsclclient clp, uint8_t fhp, int fhlen,
4901	uint64_t off, struct nfsclflayout *retflpp, int recalledp)
4902	{
4903	struct nfscllayout *lyp;
4904	mount_t mp;
4905	int error, igotlock;
4906
4907	mp = clp->nfsc_nmp->nm_mountp;
4908	*recalledp = 0;
4909	*retflpp = NULL;
4910	NFSLOCKCLSTATE();
4911	lyp = nfscl_findlayout(clp, fhp, fhlen);
4912	if (lyp != NULL) {
4913	if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
4914	TAILQ_REMOVE(&clp->nfsc_layout, lyp, nfsly_list);
4915	TAILQ_INSERT_HEAD(&clp->nfsc_layout, lyp, nfsly_list);
4916	lyp->nfsly_timestamp = NFSD_MONOSEC + 120;
4917	error = nfscl_findlayoutforio(lyp, off,
4918	NFSV4OPEN_ACCESSREAD, retflpp);
4919	if (error == 0)
4920	nfsv4_getref(&lyp->nfsly_lock, NULL,
4921	NFSCLSTATEMUTEXPTR, mp);
4922	else {
4923	do {
4924	igotlock = nfsv4_lock(&lyp->nfsly_lock,
4925	1, NULL, NFSCLSTATEMUTEXPTR, mp);
4926	} while (igotlock == 0 && !NFSCL_FORCEDISM(mp));
4927	*retflpp = NULL;
4928	}
4929	if (NFSCL_FORCEDISM(mp)) {
4930	lyp = NULL;
4931	*recalledp = 1;
4932	}
4933	} else {
4934	lyp = NULL;
4935	*recalledp = 1;
4936	}
4937	}
4938	NFSUNLOCKCLSTATE();
4939	return (lyp);
4940	}
4941
4942	/*
4943	* Search for a layout by MDS file handle. If one is found, mark in to be
4944	* recalled, if it already marked "return on close".
4945	*/
4946	static void
4947	nfscl_retoncloselayout(vnode_t vp, struct nfsclclient clp, uint8_t fhp,
4948	int fhlen, struct nfsclrecalllayout **recallpp)
4949	{
4950	struct nfscllayout *lyp;
4951	uint32_t iomode;
4952
4953	if (vp->v_type != VREG \|\| !NFSHASPNFS(VFSTONFS(vnode_mount(vp))) \|\|
4954	nfscl_enablecallb == 0 \|\| nfs_numnfscbd == 0 \|\|
4955	(VTONFS(vp)->n_flag & NNOLAYOUT) != 0)
4956	return;
4957	lyp = nfscl_findlayout(clp, fhp, fhlen);
4958	if (lyp != NULL && (lyp->nfsly_flags & (NFSLY_RETONCLOSE \|
4959	NFSLY_RECALL)) == NFSLY_RETONCLOSE) {
4960	iomode = 0;
4961	if (!LIST_EMPTY(&lyp->nfsly_flayread))
4962	iomode \|= NFSLAYOUTIOMODE_READ;
4963	if (!LIST_EMPTY(&lyp->nfsly_flayrw))
4964	iomode \|= NFSLAYOUTIOMODE_RW;
4965	(void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
4966	0, UINT64_MAX, lyp->nfsly_stateid.seqid, 0, 0, NULL,
4967	*recallpp);
4968	NFSCL_DEBUG(4, "retoncls recall iomode=%d\n", iomode);
4969	*recallpp = NULL;
4970	}
4971	}
4972
4973	/*
4974	* Mark the layout to be recalled and with an error.
4975	* Also, disable the dsp from further use.
4976	*/
4977	void
4978	nfscl_dserr(uint32_t op, uint32_t stat, struct nfscldevinfo *dp,
4979	struct nfscllayout lyp, struct nfsclds dsp)
4980	{
4981	struct nfsclrecalllayout *recallp;
4982	uint32_t iomode;
4983
4984	printf("DS being disabled, error=%d\n", stat);
4985	/* Set up the return of the layout. */
4986	recallp = malloc(sizeof(*recallp), M_NFSLAYRECALL, M_WAITOK);
4987	iomode = 0;
4988	NFSLOCKCLSTATE();
4989	if ((lyp->nfsly_flags & NFSLY_RECALL) == 0) {
4990	if (!LIST_EMPTY(&lyp->nfsly_flayread))
4991	iomode \|= NFSLAYOUTIOMODE_READ;
4992	if (!LIST_EMPTY(&lyp->nfsly_flayrw))
4993	iomode \|= NFSLAYOUTIOMODE_RW;
4994	(void)nfscl_layoutrecall(NFSLAYOUTRETURN_FILE, lyp, iomode,
4995	0, UINT64_MAX, lyp->nfsly_stateid.seqid, stat, op,
4996	dp->nfsdi_deviceid, recallp);
4997	NFSUNLOCKCLSTATE();
4998	NFSCL_DEBUG(4, "nfscl_dserr recall iomode=%d\n", iomode);
4999	} else {
5000	NFSUNLOCKCLSTATE();
5001	free(recallp, M_NFSLAYRECALL);
5002	}
5003
5004	/* And shut the TCP connection down. */
5005	nfscl_cancelreqs(dsp);
5006	}
5007
5008	/*
5009	* Cancel all RPCs for this "dsp" by closing the connection.
5010	* Also, mark the session as defunct.
5011	* If NFSCLDS_SAMECONN is set, the connection is shared with other DSs and
5012	* cannot be shut down.
5013	*/
5014	APPLESTATIC void
5015	nfscl_cancelreqs(struct nfsclds *dsp)
5016	{
5017	struct __rpc_client *cl;
5018	static int non_event;
5019
5020	NFSLOCKDS(dsp);
5021	if ((dsp->nfsclds_flags & (NFSCLDS_CLOSED \| NFSCLDS_SAMECONN)) == 0 &&
5022	dsp->nfsclds_sockp != NULL &&
5023	dsp->nfsclds_sockp->nr_client != NULL) {
5024	dsp->nfsclds_flags \|= NFSCLDS_CLOSED;
5025	cl = dsp->nfsclds_sockp->nr_client;
5026	dsp->nfsclds_sess.nfsess_defunct = 1;
5027	NFSUNLOCKDS(dsp);
5028	CLNT_CLOSE(cl);
5029	/*
5030	* This 1sec sleep is done to reduce the number of reconnect
5031	* attempts made on the DS while it has failed.
5032	*/
5033	tsleep(&non_event, PVFS, "ndscls", hz);
5034	return;
5035	}
5036	NFSUNLOCKDS(dsp);
5037	}
5038
5039	/*
5040	* Dereference a layout.
5041	*/
5042	void
5043	nfscl_rellayout(struct nfscllayout *lyp, int exclocked)
5044	{
5045
5046	NFSLOCKCLSTATE();
5047	if (exclocked != 0)
5048	nfsv4_unlock(&lyp->nfsly_lock, 0);
5049	else
5050	nfsv4_relref(&lyp->nfsly_lock);
5051	NFSUNLOCKCLSTATE();
5052	}
5053
5054	/*
5055	* Search for a devinfo by deviceid. If one is found, return it after
5056	* acquiring a reference count on it.
5057	*/
5058	struct nfscldevinfo *
5059	nfscl_getdevinfo(struct nfsclclient clp, uint8_t deviceid,
5060	struct nfscldevinfo *dip)
5061	{
5062
5063	NFSLOCKCLSTATE();
5064	if (dip == NULL)
5065	dip = nfscl_finddevinfo(clp, deviceid);
5066	if (dip != NULL)
5067	dip->nfsdi_refcnt++;
5068	NFSUNLOCKCLSTATE();
5069	return (dip);
5070	}
5071
5072	/*
5073	* Dereference a devinfo structure.
5074	*/
5075	static void
5076	nfscl_reldevinfo_locked(struct nfscldevinfo *dip)
5077	{
5078
5079	dip->nfsdi_refcnt--;
5080	if (dip->nfsdi_refcnt == 0)
5081	wakeup(&dip->nfsdi_refcnt);
5082	}
5083
5084	/*
5085	* Dereference a devinfo structure.
5086	*/
5087	void
5088	nfscl_reldevinfo(struct nfscldevinfo *dip)
5089	{
5090
5091	NFSLOCKCLSTATE();
5092	nfscl_reldevinfo_locked(dip);
5093	NFSUNLOCKCLSTATE();
5094	}
5095
5096	/*
5097	* Find a layout for this file handle. Return NULL upon failure.
5098	*/
5099	static struct nfscllayout *
5100	nfscl_findlayout(struct nfsclclient clp, u_int8_t fhp, int fhlen)
5101	{
5102	struct nfscllayout *lyp;
5103
5104	LIST_FOREACH(lyp, NFSCLLAYOUTHASH(clp, fhp, fhlen), nfsly_hash)
5105	if (lyp->nfsly_fhlen == fhlen &&
5106	!NFSBCMP(lyp->nfsly_fh, fhp, fhlen))
5107	break;
5108	return (lyp);
5109	}
5110
5111	/*
5112	* Find a devinfo for this deviceid. Return NULL upon failure.
5113	*/
5114	static struct nfscldevinfo *
5115	nfscl_finddevinfo(struct nfsclclient clp, uint8_t deviceid)
5116	{
5117	struct nfscldevinfo *dip;
5118
5119	LIST_FOREACH(dip, &clp->nfsc_devinfo, nfsdi_list)
5120	if (NFSBCMP(dip->nfsdi_deviceid, deviceid, NFSX_V4DEVICEID)
5121	== 0)
5122	break;
5123	return (dip);
5124	}
5125
5126	/*
5127	* Merge the new file layout list into the main one, maintaining it in
5128	* increasing offset order.
5129	*/
5130	static void
5131	nfscl_mergeflayouts(struct nfsclflayouthead *fhlp,
5132	struct nfsclflayouthead *newfhlp)
5133	{
5134	struct nfsclflayout flp, nflp, prevflp, tflp;
5135
5136	flp = LIST_FIRST(fhlp);
5137	prevflp = NULL;
5138	LIST_FOREACH_SAFE(nflp, newfhlp, nfsfl_list, tflp) {
5139	while (flp != NULL && flp->nfsfl_off < nflp->nfsfl_off) {
5140	prevflp = flp;
5141	flp = LIST_NEXT(flp, nfsfl_list);
5142	}
5143	if (prevflp == NULL)
5144	LIST_INSERT_HEAD(fhlp, nflp, nfsfl_list);
5145	else
5146	LIST_INSERT_AFTER(prevflp, nflp, nfsfl_list);
5147	prevflp = nflp;
5148	}
5149	}
5150
5151	/*
5152	* Add this nfscldevinfo to the client, if it doesn't already exist.
5153	* This function consumes the structure pointed at by dip, if not NULL.
5154	*/
5155	APPLESTATIC int
5156	nfscl_adddevinfo(struct nfsmount nmp, struct nfscldevinfo dip, int ind,
5157	struct nfsclflayout *flp)
5158	{
5159	struct nfsclclient *clp;
5160	struct nfscldevinfo *tdip;
5161	uint8_t *dev;
5162
5163	NFSLOCKCLSTATE();
5164	clp = nmp->nm_clp;
5165	if (clp == NULL) {
5166	NFSUNLOCKCLSTATE();
5167	if (dip != NULL)
5168	free(dip, M_NFSDEVINFO);
5169	return (ENODEV);
5170	}
5171	if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5172	dev = flp->nfsfl_dev;
5173	else
5174	dev = flp->nfsfl_ffm[ind].dev;
5175	tdip = nfscl_finddevinfo(clp, dev);
5176	if (tdip != NULL) {
5177	tdip->nfsdi_layoutrefs++;
5178	if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5179	flp->nfsfl_devp = tdip;
5180	else
5181	flp->nfsfl_ffm[ind].devp = tdip;
5182	nfscl_reldevinfo_locked(tdip);
5183	NFSUNLOCKCLSTATE();
5184	if (dip != NULL)
5185	free(dip, M_NFSDEVINFO);
5186	return (0);
5187	}
5188	if (dip != NULL) {
5189	LIST_INSERT_HEAD(&clp->nfsc_devinfo, dip, nfsdi_list);
5190	dip->nfsdi_layoutrefs = 1;
5191	if ((flp->nfsfl_flags & NFSFL_FILE) != 0)
5192	flp->nfsfl_devp = dip;
5193	else
5194	flp->nfsfl_ffm[ind].devp = dip;
5195	}
5196	NFSUNLOCKCLSTATE();
5197	if (dip == NULL)
5198	return (ENODEV);
5199	return (0);
5200	}
5201
5202	/*
5203	* Free up a layout structure and associated file layout structure(s).
5204	*/
5205	APPLESTATIC void
5206	nfscl_freelayout(struct nfscllayout *layp)
5207	{
5208	struct nfsclflayout flp, nflp;
5209	struct nfsclrecalllayout rp, nrp;
5210
5211	LIST_FOREACH_SAFE(flp, &layp->nfsly_flayread, nfsfl_list, nflp) {
5212	LIST_REMOVE(flp, nfsfl_list);
5213	nfscl_freeflayout(flp);
5214	}
5215	LIST_FOREACH_SAFE(flp, &layp->nfsly_flayrw, nfsfl_list, nflp) {
5216	LIST_REMOVE(flp, nfsfl_list);
5217	nfscl_freeflayout(flp);
5218	}
5219	LIST_FOREACH_SAFE(rp, &layp->nfsly_recall, nfsrecly_list, nrp) {
5220	LIST_REMOVE(rp, nfsrecly_list);
5221	free(rp, M_NFSLAYRECALL);
5222	}
5223	nfscl_layoutcnt--;
5224	free(layp, M_NFSLAYOUT);
5225	}
5226
5227	/*
5228	* Free up a file layout structure.
5229	*/
5230	APPLESTATIC void
5231	nfscl_freeflayout(struct nfsclflayout *flp)
5232	{
5233	int i, j;
5234
5235	if ((flp->nfsfl_flags & NFSFL_FILE) != 0) {
5236	for (i = 0; i < flp->nfsfl_fhcnt; i++)
5237	free(flp->nfsfl_fh[i], M_NFSFH);
5238	if (flp->nfsfl_devp != NULL)
5239	flp->nfsfl_devp->nfsdi_layoutrefs--;
5240	}
5241	if ((flp->nfsfl_flags & NFSFL_FLEXFILE) != 0)
5242	for (i = 0; i < flp->nfsfl_mirrorcnt; i++) {
5243	for (j = 0; j < flp->nfsfl_ffm[i].fhcnt; j++)
5244	free(flp->nfsfl_ffm[i].fh[j], M_NFSFH);
5245	if (flp->nfsfl_ffm[i].devp != NULL)
5246	flp->nfsfl_ffm[i].devp->nfsdi_layoutrefs--;
5247	}
5248	free(flp, M_NFSFLAYOUT);
5249	}
5250
5251	/*
5252	* Free up a file layout devinfo structure.
5253	*/
5254	APPLESTATIC void
5255	nfscl_freedevinfo(struct nfscldevinfo *dip)
5256	{
5257
5258	free(dip, M_NFSDEVINFO);
5259	}
5260
5261	/*
5262	* Mark any layouts that match as recalled.
5263	*/
5264	static int
5265	nfscl_layoutrecall(int recalltype, struct nfscllayout *lyp, uint32_t iomode,
5266	uint64_t off, uint64_t len, uint32_t stateseqid, uint32_t stat, uint32_t op,
5267	char devid, struct nfsclrecalllayout recallp)
5268	{
5269	struct nfsclrecalllayout rp, orp;
5270
5271	recallp->nfsrecly_recalltype = recalltype;
5272	recallp->nfsrecly_iomode = iomode;
5273	recallp->nfsrecly_stateseqid = stateseqid;
5274	recallp->nfsrecly_off = off;
5275	recallp->nfsrecly_len = len;
5276	recallp->nfsrecly_stat = stat;
5277	recallp->nfsrecly_op = op;
5278	if (devid != NULL)
5279	NFSBCOPY(devid, recallp->nfsrecly_devid, NFSX_V4DEVICEID);
5280	/*
5281	* Order the list as file returns first, followed by fsid and any
5282	* returns, both in increasing stateseqid order.
5283	* Note that the seqids wrap around, so 1 is after 0xffffffff.
5284	* (I'm not sure this is correct because I find RFC5661 confusing
5285	* on this, but hopefully it will work ok.)
5286	*/
5287	orp = NULL;
5288	LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5289	orp = rp;
5290	if ((recalltype == NFSLAYOUTRETURN_FILE &&
5291	(rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE \|\|
5292	nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) \|\|
5293	(recalltype != NFSLAYOUTRETURN_FILE &&
5294	rp->nfsrecly_recalltype != NFSLAYOUTRETURN_FILE &&
5295	nfscl_seq(stateseqid, rp->nfsrecly_stateseqid) != 0)) {
5296	LIST_INSERT_BEFORE(rp, recallp, nfsrecly_list);
5297	break;
5298	}
5299
5300	/*
5301	* Put any error return on all the file returns that will
5302	* preceed this one.
5303	*/
5304	if (rp->nfsrecly_recalltype == NFSLAYOUTRETURN_FILE &&
5305	stat != 0 && rp->nfsrecly_stat == 0) {
5306	rp->nfsrecly_stat = stat;
5307	rp->nfsrecly_op = op;
5308	if (devid != NULL)
5309	NFSBCOPY(devid, rp->nfsrecly_devid,
5310	NFSX_V4DEVICEID);
5311	}
5312	}
5313	if (rp == NULL) {
5314	if (orp == NULL)
5315	LIST_INSERT_HEAD(&lyp->nfsly_recall, recallp,
5316	nfsrecly_list);
5317	else
5318	LIST_INSERT_AFTER(orp, recallp, nfsrecly_list);
5319	}
5320	lyp->nfsly_flags \|= NFSLY_RECALL;
5321	wakeup(lyp->nfsly_clp);
5322	return (0);
5323	}
5324
5325	/*
5326	* Compare the two seqids for ordering. The trick is that the seqids can
5327	* wrap around from 0xffffffff->0, so check for the cases where one
5328	* has wrapped around.
5329	* Return 1 if seqid1 comes before seqid2, 0 otherwise.
5330	*/
5331	static int
5332	nfscl_seq(uint32_t seqid1, uint32_t seqid2)
5333	{
5334
5335	if (seqid2 > seqid1 && (seqid2 - seqid1) >= 0x7fffffff)
5336	/* seqid2 has wrapped around. */
5337	return (0);
5338	if (seqid1 > seqid2 && (seqid1 - seqid2) >= 0x7fffffff)
5339	/* seqid1 has wrapped around. */
5340	return (1);
5341	if (seqid1 <= seqid2)
5342	return (1);
5343	return (0);
5344	}
5345
5346	/*
5347	* Do a layout return for each of the recalls.
5348	*/
5349	static void
5350	nfscl_layoutreturn(struct nfsmount nmp, struct nfscllayout lyp,
5351	struct ucred cred, NFSPROC_T p)
5352	{
5353	struct nfsclrecalllayout *rp;
5354	nfsv4stateid_t stateid;
5355	int layouttype;
5356
5357	NFSBCOPY(lyp->nfsly_stateid.other, stateid.other, NFSX_STATEIDOTHER);
5358	stateid.seqid = lyp->nfsly_stateid.seqid;
5359	if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5360	layouttype = NFSLAYOUT_NFSV4_1_FILES;
5361	else
5362	layouttype = NFSLAYOUT_FLEXFILE;
5363	LIST_FOREACH(rp, &lyp->nfsly_recall, nfsrecly_list) {
5364	(void)nfsrpc_layoutreturn(nmp, lyp->nfsly_fh,
5365	lyp->nfsly_fhlen, 0, layouttype,
5366	rp->nfsrecly_iomode, rp->nfsrecly_recalltype,
5367	rp->nfsrecly_off, rp->nfsrecly_len,
5368	&stateid, cred, p, rp->nfsrecly_stat, rp->nfsrecly_op,
5369	rp->nfsrecly_devid);
5370	}
5371	}
5372
5373	/*
5374	* Do the layout commit for a file layout.
5375	*/
5376	static void
5377	nfscl_dolayoutcommit(struct nfsmount nmp, struct nfscllayout lyp,
5378	struct ucred cred, NFSPROC_T p)
5379	{
5380	struct nfsclflayout *flp;
5381	uint64_t len;
5382	int error, layouttype;
5383
5384	if ((lyp->nfsly_flags & NFSLY_FILES) != 0)
5385	layouttype = NFSLAYOUT_NFSV4_1_FILES;
5386	else
5387	layouttype = NFSLAYOUT_FLEXFILE;
5388	LIST_FOREACH(flp, &lyp->nfsly_flayrw, nfsfl_list) {
5389	if (layouttype == NFSLAYOUT_FLEXFILE &&
5390	(flp->nfsfl_fflags & NFSFLEXFLAG_NO_LAYOUTCOMMIT) != 0) {
5391	NFSCL_DEBUG(4, "Flex file: no layoutcommit\n");
5392	/* If not supported, don't bother doing it. */
5393	NFSLOCKMNT(nmp);
5394	nmp->nm_state \|= NFSSTA_NOLAYOUTCOMMIT;
5395	NFSUNLOCKMNT(nmp);
5396	break;
5397	} else if (flp->nfsfl_off <= lyp->nfsly_lastbyte) {
5398	len = flp->nfsfl_end - flp->nfsfl_off;
5399	error = nfsrpc_layoutcommit(nmp, lyp->nfsly_fh,
5400	lyp->nfsly_fhlen, 0, flp->nfsfl_off, len,
5401	lyp->nfsly_lastbyte, &lyp->nfsly_stateid,
5402	layouttype, cred, p, NULL);
5403	NFSCL_DEBUG(4, "layoutcommit err=%d\n", error);
5404	if (error == NFSERR_NOTSUPP) {
5405	/* If not supported, don't bother doing it. */
5406	NFSLOCKMNT(nmp);
5407	nmp->nm_state \|= NFSSTA_NOLAYOUTCOMMIT;
5408	NFSUNLOCKMNT(nmp);
5409	break;
5410	}
5411	}
5412	}
5413	}
5414
5415	/*
5416	* Commit all layouts for a file (vnode).
5417	*/
5418	int
5419	nfscl_layoutcommit(vnode_t vp, NFSPROC_T *p)
5420	{
5421	struct nfsclclient *clp;
5422	struct nfscllayout *lyp;
5423	struct nfsnode *np = VTONFS(vp);
5424	mount_t mp;
5425	struct nfsmount *nmp;
5426
5427	mp = vnode_mount(vp);
5428	nmp = VFSTONFS(mp);
5429	if (NFSHASNOLAYOUTCOMMIT(nmp))
5430	return (0);
5431	NFSLOCKCLSTATE();
5432	clp = nmp->nm_clp;
5433	if (clp == NULL) {
5434	NFSUNLOCKCLSTATE();
5435	return (EPERM);
5436	}
5437	lyp = nfscl_findlayout(clp, np->n_fhp->nfh_fh, np->n_fhp->nfh_len);
5438	if (lyp == NULL) {
5439	NFSUNLOCKCLSTATE();
5440	return (EPERM);
5441	}
5442	nfsv4_getref(&lyp->nfsly_lock, NULL, NFSCLSTATEMUTEXPTR, mp);
5443	if (NFSCL_FORCEDISM(mp)) {
5444	NFSUNLOCKCLSTATE();
5445	return (EPERM);
5446	}
5447	tryagain:
5448	if ((lyp->nfsly_flags & NFSLY_WRITTEN) != 0) {
5449	lyp->nfsly_flags &= ~NFSLY_WRITTEN;
5450	NFSUNLOCKCLSTATE();
5451	NFSCL_DEBUG(4, "do layoutcommit2\n");
5452	nfscl_dolayoutcommit(clp->nfsc_nmp, lyp, NFSPROCCRED(p), p);
5453	NFSLOCKCLSTATE();
5454	goto tryagain;
5455	}
5456	nfsv4_relref(&lyp->nfsly_lock);
5457	NFSUNLOCKCLSTATE();
5458	return (0);
5459	}
5460

Note: See TracBrowser for help on using the repository browser.

Download in other formats: